**An Optical Fiber Sensor for Hg2+ Detection Based on the LSPR of Silver and Gold Nanoparticles Embedded in a Polymeric Matrix as an Effective Sensing Material †**

**María Elena Martínez-Hernández 1,\* , Xabier Sandua 2,3 , Pedro J. Rivero 2,3, Javier Goicoechea 1,4 and Francisco J. Arregui 1,4**


**Abstract:** In this work, an optical fiber sensor based on the localized surface plasmon resonance (LSPR) phenomenon is presented as a powerful tool for the detection of heavy metals (Hg2+). The resultant sensing film was fabricated using a nanofabrication process, known as layer-by-layer embedding (LbL-E) deposition technique. In this sense, both silver nanoparticles (AgNPs) and gold nanoparticles (AuNPs) were synthesized using a synthetic chemical protocol as a function of a strict control of three main parameters: polyelectrolyte concentration, loading agent, and reducing agent. The use of metallic nanostructures as sensing materials is of great interest because well-located absorption peaks associated with their LSPR are obtained at 420 nm (AgNPs) and 530 nm (AuNPs). Both plasmonic peaks provide a stable real-time reference that can be extracted from the spectral response of the optical fiber sensor, giving a reliable monitoring of the Hg2+ concentration.

**Keywords:** fiber optic sensor; gold nanoparticles; silver nanoparticles; localized surface plasmon resonance; layer-by-layer embedding; mercury; ppm

#### **1. Introduction**

The presence of heavy metals in a human's daily life has become a concern due to their adverse health effects. Among all of them, mercury is receiving major attention because its presence is associated with serious problems, such as lung or nervous system damage, heart diseases, and even neurological and psychological symptoms [1]. Due to this, a wide variety of detection methods can be found in the bibliography, ranging from electrochemical sensors [2–4] to colorimetric sensors [5–7] and optical sensors [8–10]. This work focuses on optical fiber sensors based on the localized surface plasmon resonance (LSPR) phenomenon. It is well known that LSPR is an optical phenomenon that is generated, thanks to the interaction between the incident light and the electrons in the conduction band of the metal surface [11]. It has been demonstrated that the resultant amplitude and the plasmonic resonance energy can vary as a function of the geometry and the distance between the nanoparticles. Until now, LSPR optical fiber sensors for mercury ion detection mostly contain gold nanoparticles (AuNPs) as the main plasmonic sensing material, showing the interaction between gold and mercury as a change in the physical and chemical properties of

**Citation:** Martínez-Hernández, M.E.; Sandua, X.; Rivero, P.J.; Goicoechea, J.; Arregui, F.J. An Optical Fiber Sensor for Hg2+ Detection Based on the LSPR of Silver and Gold Nanoparticles Embedded in a Polymeric Matrix as an Effective Sensing Material. *Chem. Proc.* **2021**, *5*, 73. https://doi.org/10.3390/ CSAC2021-10633

Academic Editor: Huangxian Ju

Published: 7 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

the metallic nanoparticles [12,13]. The novelty of this work is the possibility of introducing two different metallic nanoparticles, AgNPs and AuNPs, into LbL films with the aim of obtaining two different LSPR sensing signals for the detection of mercury ions. This deposition technique makes it possible to obtain thin films with a good control in the resultant thickness in the nanometric range as a function of operational parameters, such as pH, ionic strength, and number of bilayers deposited [14–16]. An initial study is performed on glass slides in order to optimize the nanofabrication technique, and then sensing coating is implemented on an optical fiber. Finally, a change in the wavelength position of the LSPR band can be observed as a function of the concentration of the analyte. To sum up, this is the first time that an optical fiber sensor with a dual reference state is presented for mercury ion detection.

#### **2. Methods**

#### *2.1. Materials*

The polymeric matrix is composed of poly(allylamine hydrochloride) (PAH) (Mw ~15.000), which acts as a polycation, and poly(acrylic acid) (PAA) 35 wt% solution in water, which acts as a polyanion. In order to obtain AuNPs and AgNPs, gold(III) chloride trihydrate (HAuCl4·3H2O) and silver nitrate (AgNO3) were used as loading agents for the synthesis of metallic nanoparticles. Finally, dimethylamine borane complex (DMAB) was used as a reducing agent.

#### *2.2. Chemical Process for the Synthesis of Metallic Nanoparticles*

#### 2.2.1. Gold Nanoparticle (AuNP) Synthesis

First, aqueous solutions of HAuCl4·3H2O (20 mL, 5 mM) and PAA (120 mL, 10 mM), which acts as a stabilizing agent, were mixed and stirred for a period of 2 h. After that, an aqueous solution of the reducing agent, DMAB (5 mL, 100 mM), was added to the previous solution, and the mixture was stirred for 24 h at room temperature. Finally, a color change from yellow to violet was obtained, indicating the synthesis of AuNPs. The combination of PAA and AuNPs is denoted as PAA-AuNPs. This colloidal dispersion solution (PAA-AuNPs) showed a spherical shape and nanometric range (10–20 nm) corroborated by transmission electron microscopy (TEM) [17].

#### 2.2.2. Silver Nanoparticle (AgNPs) Synthesis

For the synthesis of AgNPs, first, aqueous solutions of AgNO3 (20 mL, 10 mM) and PAH (120 mL, 10 mM), which acts as a stabilizing agent, were mixed and stirred for a period of 2 h. After that, an aqueous solution of the reducing agent, DMAB (5 mL, 100 mM), was added to the initial solution, and the mixture was stirred for 24 h at room temperature. Finally, a color change from transparent to orange was obtained, indicating the synthesis of AgNPs. The combination of PAH and AgNPs is denoted as PAH-AgNPs. The location of the LSPR absorption band clearly indicates the synthesis of AgNPs with a spherical shape and nanometric size [18].

#### *2.3. Optical Characterization*

The optical properties of the synthesized metallic nanoparticles were determined by using a Jasco V-630 spectrophotometer (Agilent, Santa Clara, CA, USA). Two different and well-separated absorption bands were obtained.

#### *2.4. Layer-by-Layer Nanoassembly*

The layer-by-layer nanoassembly technique was used for the fabrication of the thin films. In this work, the presence of PAH and PAA was used as the positively and negatively charged polyelectrolytes for the buildup of the polyelectrolyte structure film. In addition, as demonstrated in the previous section, these charged structures also play a key role in stabilizing the synthesized nanoparticles. More specifically, the polycationic solution PAH-

capped AgNPs (PAH-AgNPs) and the polyanionic PAA-capped AuNPs (PAA-AuNPs) were used for the fabrication of the thin films.

#### *2.5. Optical Fiber Detection Setup*

A multimode optical fiber with a 200 μm core diameter with polymeric cladding, 0.39 NA (Thorlabs FT-200-EMT), was used for the fabrication of the optical fiber sensor. First, it was necessary to remove the acrylate cladding of a segment of approximately 2 cm of the optical fiber, and for this, a few drops of dry acetone and a blade were used, exposing the bare optical fiber core in its entire cylindrical section. This optical fiber segment was immersed for 5 min in piranha solution to eliminate the acetone that could remain. Temporary SMA connectors were used at the end of the optical fiber, exciting the sensor from one of the connectors with a halogen white source, and the other end collected the optical response with a CCD spectrometer (HR4000-UV Ocean Optics, Ocean Insight, FL, USA). A scheme of the deposition process and the optical setup is presented in Figure 1.

**Figure 1.** Schematic representation for the fabrication of the LbL films by using PAH-AgNPs as a polycation and PAA-AuNPs as a polyanion.

#### *2.6. Mercury Sample Preparation*

The mercury samples were prepared using mercury(II) chloride (HgCl2). Every concentration of mercury was prepared with phosphate buffer (PB) solution for achieving a constant pH = 7.6. The Hg concentrations were varied from 50 to 1 and 0.1 ppm. An important aspect is that for each measurement, the fiber optical sensor was immersed in PB + DMAB buffer solution with the aim of obtaining a stable baseline for further mercury detection.

#### **3. Results and Discussion**

As an initial step, the nanofabrication process was performed on glass substrates, and then, the same procedure was extrapolated to the optical fiber for further chemical sensing. The selected pH for the fabrication of the whole process was 7.0 in the dipping polyelectrolytes. As can be observed in Figure 2, the sample for thickness coating with 10 bilayers showed a clear predominance of the LSPR in relation to AgNPs (plasmonic peak centered at 450 nm), without being able to identify the peak in relation to AuNPs. However, when the thickness coating was gradually increased up to a total thickness of 30 bilayers, both LSPR peaks can be clearly observed, which were centered at 420 nm (AgNPs) and 540 nm (AuNPs), although transparent films were still obtained, which were observed by the naked eye.

**Figure 2.** UV–VIS spectra of the LbL coatings based on PAH/AgNPs and PAA/AuNPs deposited on glass slides as a function of the thickness coating (10, 20, and 30 bilayers) for pH 7.0.

Once the presence of both LSPR peaks was demonstrated on glass slides, the next step was based on the deposition of this same thin film onto an optical fiber at the same pH value (87.0) in order to appreciate both absorption bands in the UV–VIS spectra. In Figure 3, it is demonstrated that by only a final thickness of 7 bilayers, it is possible to appreciate the LSPR of the AgNPs (centered at 420 nm) and AuNPs (centered at 540 nm), this sensing thin film being for the mercury ion detection.

**Figure 3.** UV–VIS spectrum of the LbL coatings based on PAH/AgNPs and PAA/AuNPs deposited on an optical fiber for a thickness coating of 7 bilayers.

#### *Detection of Mercury Ions with Fiber Optic Sensor*

Once the thin film was fabricated, the optical fiber was immersed in the Buffer PB + DMAB solution for 1 h in order to have a stable baseline for the mercury detection stage. After that, the sensing film was immersed at a fixed mercury concentration of 50 ppm, and a very interesting result was that a clear wavelength shift of 23 nm was observed for LSPR (AuNPs), which remained stable in wavelength, compared with the wavelength shift observed in the AuNP LSPR band. According to this, LSPR-AuNPs are much more sensitive to the presence of Hg than LSPR-AgNPs (Figure 4). The UV–VIS spectra for the minimum (0.1 ppm) and maximum (50 ppm) H2O2 concentrations are presented in Figure 5 in order to have a better appreciation of the wavelength shift related to the LSPR-AuNPs, with the detection range being in the order of 100 ppb.

**Figure 4.** Wavelength shift of the LSPR absorption bands at 50 ppm of mercury concentration.

**Figure 5.** UV–VIS spectra for (**a**) the minimum concentration (0.1 ppm) and (**b**) the maximum Hg concentration (50 ppm).

The wavelength shift observed in the AuNP LSPR absorption band can be explained by the chemical reaction of the mercury present in the sample with the AuNPs. As previously reported [8], mercury ions, in the presence of DMAB of the stock buffered solution, are reduced to metallic mercury, which is known to show a high affinity for gold to form an amalgam [13]. Mercury reacts with AuNPs, changing their surface chemistry, so it is possible that their effective diameter in terms of LSPR resonances is reduced, which explains the blue shift in the LSPR band maximum. The higher reaction affinity of mercury towards gold compared with silver makes the stability of the AgNP absorption band possible, which can be used as a quite stable wavelength reference.

Different sensors were fabricated with the same sensing materials in order to detect a particular mercury concentration. Although both LSPR bands experimented changes in the presence of mercury ions, it is clearly visible that the LSPR band corresponding to AuNPs showed a greater blue shift in comparison with the LSPR of AgNPs. Finally, the dynamic response of the LSPR band inherent in AuNPs is presented in Figure 6 for different mercury concentrations (0.1, 1, and 50 ppm).

**Figure 6.** Dynamic response of the optical fiber sensors for the LSPR (AuNPs) to different Hg concentrations, ranging from 50 to 0.1 ppm.

#### **4. Conclusions**

In this work, a fiber optic sensor based on two different LSPR sensing signals for the detection of Hg2+ was presented. The metallic nanoparticles were incorporated into the sensing films by using the layer-by-layer nanoassembly technique. The sensors were exposed to different Hg2+ concentrations, with the wavelength response of the AuNP LSPR greater than that of the AgNP LSPR. Finally, this resultant sensing material can be extrapolated for the detection of different heavy metals in environmental applications.

**Author Contributions:** Conceptualization and methodology, M.E.M.-H., X.S., P.J.R., J.G. and F.J.A.; investigation and validation, M.E.M.-H. and X.S.; writing—original draft preparation, M.E.M.-H., X.S. and P.J.R.; writing—review and editing, P.J.R., J.G. and F.J.A.; supervision, P.J.R., J.G. and F.J.A.; project administration and funding acquisition, F.J.A. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work has been supported by the Spanish State Research Agency (AEI) through the project PID2019-106070RB-I00 and the Public University of Navarre with a PhD research grant.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Proceeding Paper* **Development of Graphene-Doped TiO2-Nanotube Array-Based MIM-Structured Sensors and Its Application for Methanol Sensing at Room Temperature †**

**Teena Gakhar \* and Arnab Hazra**

Department of Electrical and Electronics Engineering, Birla Institute of Technology and Science BITS, Pilani 333031, India; arnabhazra2013@gmail.com

**\*** Correspondence: gakharteena16@gmail.com

† Presented at the 1st International Electronic Conference on Chemical Sensors and Analytical Chemistry, 1–15 July 2021; Available online: https://csac2021.sciforum.net/.

**Abstract:** This work concerns the development of a good quality graphene doped TiO2 nanotube array sensor for efficient detection of methanol. A pure and graphene doped TiO2 nanotube array was synthesized by electrochemical anodization. Morphological, structural and optical characterizations were performed to study the samples. Both the nanotube samples were produced in Au/TiO2 nanotube/Ti type MIM-structured devices. Pure and graphene-doped TiO2 nanotubes offered a response magnitude of 20% and 28% to 100 ppm of methanol at room temperature, respectively. Response/Recovery time was fast for the graphene doped TiO2 nanotube array (34 s/40 s) compared to a pure TiO2 nanotube array (116 s/576 s) at room temperature. This study confirmed the notable enhancement in methanol sensing due to the formation of local heterojunctions between graphene and TiO2 in the hybrid sample.

**Keywords:** methanol sensing; graphene doping; electrochemical anodization

**Citation:** Gakhar, T.; Hazra, A. Development of Graphene-Doped TiO2-Nanotube Array-Based MIM-Structured Sensors and Its Application for Methanol Sensing at Room Temperature. *Chem. Proc.* **2021**, *5*, 74. https://doi.org/10.3390/ CSAC2021-10620

Academic Editor: Ye Zhou

Published: 6 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

#### **1. Introduction**

Methanol is one of the essential organic solvents having numerous applications in the production of dyes, drugs, perfumes and colors. Moreover, it is extensively utilized in automobile fuel, wastewater denitrification and electricity generation [1]. Methanol is an extremely toxic VOC which is disastrous to human health. Repeated exposure to methanol vapors causes many problems to human beings, such as blindness, acidosis, headaches, blurred vision, shortness of breath and dizziness. Skin contact with methanol results in dermatitis or scaling and eye contact results in vision destruction [2]. With all these concerns, there is a high demand for the development of methanol sensors which are reliable, stable, and sensitive as well as able to perform at low temperatures.

Different materials like metal oxide semiconductors, polymers, carbon nanostructures, metal nanoparticles, and nanocomposites have been extensively utilized by different researchers for chemical sensing. Solid state sensors based on semiconducting metal oxides have achieved a lot in the field of chemical sensing due to their exceptional properties [3]. TiO2 is an efficient semiconducting metal oxide which can be synthesized in different nanoforms (nanotubes, nanorods, nanoparticles and nanospheres, etc.) for different applications like photocatalysis [4], chemical sensing [5], and wastewater purification [6]. In the field of vapor sensing, the 1D-TiO2 nanotube performs very well due to its ideal properties such as uniformity, stability and one dimensional electron flow [7]. Different researchers have applied different techniques such as the formation of TiO2-based hybrid to improvise the performance of TiO2-based sensors.

Two-dimensional graphene offers advanced opportunities to develop hybrids with amazing electronic catalytic behavior. The flat monolayer of graphene offers unique properties such as high surface-to-volume ratio, excessive mobility and good electrical conductivity [8]. These properties make graphene an ideal candidate to support or form hybrid with metal oxide semiconductors having high catalytic properties [9]. Some reports have been published demonstrating the sensing performance of a graphene-TiO2-based hybrid. Fan and group described the hydrothermal production of a TiO2-graphene nanocomposite and its implementation in electrochemical sensing. They showed electrochemical sensing of dopamine with excellent sensitivity and selectivity [10]. Ye and co-workers reported room temperature ammonia sensing by an rGO-TiO2 hybrid. They produced the hybrid by a simple hydrothermal method [11]. Galstyan and group reported the production of an rGO-TiO2 nanotube hybrid for hydrogen sensing. They showed the impact of GO concentration on the response of TiO2 nanotubes [12].

In this current work, a highly aligned and uniform graphene-doped TiO2 nanotube array was synthesized by way of electrochemical anodization for efficient detection of methanol vapors. A pure TiO2 nanotube array and graphene-doped TiO2 nanotube array were produced by way of electrochemical anodization. Both samples were examined and analyzed through various characterization techniques which confirmed the presence of graphene in the graphene-doped TiO2 nanotube array. Metal-insulator-metal (MIM) structured sensors were produced by using both pure and graphene-doped TiO2 nanotubes. Graphene-doped TiO2 nanotubes showed a sensitivity of 28% with quite a fast response and recovery time of 34 s and 40 s towards 100 ppm of methanol. A pure TiO2 nanotube array, however, showed a sensitivity of 20% with relatively slow response/recovery time (116 s/576 s) in the same conditions.

#### **2. Experimental Details**

A highly ordered and oriented pure TiO2 nanotube array and graphene-doped TiO2 nanotube array were synthesized by electrochemical anodization route. Two-electrode anodic oxidation was performed for 120 min under 40 V potential where Ti foil was used as the anode and graphite was used as the cathode. The electrolyte was made up of 0.5 wt% of NH4F, 10% vol of DI water and ethylene glycol. The method to synthesize the TiO2 nanotube array was described in detail in our previous reports [13].

High purity graphene oxide suspension was used to prepare 0.2 wt% graphene oxide (GO) aqueous solution. Then an electrolyte was prepared with 0.5 wt% NH4F, 10 vol% of GO aqueous solution and ethylene glycol for the preparation of a graphene-doped TiO2 nanotube array. Again, the anodization was performed for 120 min by applying a constant voltage of 40 V. Due to the constant availability of GO in the electrolyte, graphene was doped uniformly in the TiO2 nanotubes. Both the pure TiO2 nanotube array and graphenedoped TiO2 nanotube array were annealed for 3 h at 450 ◦C in ambient air. Annealing made the nanotubes more robust and stable and hence more reliable to use. The flow chart describing the steps for the synthesis of pure TiO2 nanotube array and graphene doped TiO2 nanotube array is represented in Figure 1.

The morphology of the produced samples was analysed by FESEM. The crystallographic structure of both the samples was examined via X-ray diffraction spectroscopy. Raman spectroscopy was performed for both samples which confirmed the doping of graphene in graphene-doped TiO2 nanotube array (GO-TiO2).

To produce the MIM structure for the sensors, Au was deposited on top of TiO2 nanotube/Ti and GO-TiO2 nanotube/Ti samples by electron beam evaporation.. 100 nm thick deposited Au was considered as the top electrode and Ti was considered as the bottom electrode. Both samples were enveloped in Cu mask to ensure 1\*1mm<sup>2</sup> Au top electrode. A part from the corner of the TiO2 nanotubes was etched with hydro fluoric acid to induce Ti as the bottom electrode.

**Figure 1.** Flow chart describing the synthesis procedure of pure TiO2 nanotube array and graphenedoped TiO2 nanotube array were synthesized.

The produced sensors were tested against the methanol vapours. The sensors were examined at room temperature. The sensor setup with their properties has been discussed previously [14]. Resistance in the ambient air (Ra) and in the exposure of the reducing vapours (Rg), methanol was observed. The response magnitude is calculated as [(Ra − Rg)/Ra] \* 100. The response time and recovery time for both sensors is defined as 90% of maximum change of the resistance when exposed to methanol vapours and exposed to synthetic air for the removal of vapors, respectively.

#### **3. Results and Discussion**

#### *3.1. Material Characterization*

FESEM confirmed the formation of highly ordered and uniform nanotubes in both the samples (Figure 2). Highly aligned nanotubes were formed with an approximate average outer diameter of 110 nm and length of 1 μm in both the pure TiO2 nanotube array and graphene doped TiO2 nanotube array. Graphene does not hamper the original morphology of TiO2 nanotubes (Figure 2c,d). As graphene was uniformly doped inside the nanotubes, it was hard to observe the graphene with scanning electron microscopy. The chemical composition was studied through the EDS spectra, where the evidence of carbon is clearly visible in a GO-doped TiO2 nanotube array (Figure 2f).

**Figure 2.** FESEM Image of Pure TiO2 nanotube array (**a**) Top view, (**b**) Side view (**e**) EDS spectra and Graphene-doped TiO2 nanotube array (**c**) Top view, (**d**) Side view (**f**) EDS spectra.

The sharp intensity peak at 25.3◦ in both the samples is attributed to the anatase crystallinity of TiO2 nanotubes (Figure 3a). A low intensity peak at 52◦ corresponds to the anatase crystallinity A (105) in both the samples. A small peak at 54.1◦, present only in the pure TiO2 nanotube array, corresponds to A (201) and clearly shows the presence of more anatase in the pure TiO2 nanotube array. The peaks labelled as T arise due to the use of a Titanium substrate in both the samples. T peak intensity is high in the pure TiO2 nanotube array and less in the graphene-doped TiO2 nanotube array in comparison to the A (101) peak.

**Figure 3.** Pure TiO2 nanotube array and Graphene doped TiO2 nanotube array (**a**) XRD Spectra, (**b**) Raman spectra.

The Raman spectra of pure TiO2 nanotube array and graphene-doped TiO2 nanotube array is represented in Figure 3b. The presence of pure anatase is determined by six active modes Eg (144 cm−1), Eg(197 cm−1), Bg (399 cm−1) Ag + Bg (516 cm−1) and Eg (639 cm−1) present in both the samples [13]. The sharp intensity peak at 144 cm-1 determines the formation of Ti-O in the anatase phase of TiO2. The presence of graphene is authenticated by the sharp peaks at 1348 cm−<sup>1</sup> (D band) and 1596 cm−<sup>1</sup> (G band) in the graphene-doped TiO2 nanotube array [15]. The active modes of the anatase TiO2 nanotube and graphene were present at their corresponding positions even after the uniform doping of graphene.

#### *3.2. Methanol Sensing*

The two MIM-structure based sensors were examined against the reducing vapours, methanol, at room temperature. The resistance of the pure TiO2 nanotube array and graphene-doped TiO2 nanotube array was 90 MΩ and 30 MΩ, respectively. The reduced resistance (increased conductance) of graphene-doped TiO2 nanotube array sensor clearly defines the incorporation of graphene inside a TiO2 nanotube. Both the sensors were subjected to 100 ppm of methanol at room temperature. The response magnitude of pure the TiO2 nanotube array and graphene-doped TiO2 nanotube array was 20% and 28%, respectively. The pure TiO2 nanotube array had a response time and recovery time of 116 s and 576 s, respectively. Moreover, there was the improvement in response time (34 s) and recovery time (40 s) in the case of the graphene-doped TiO2 nanotube array (Figure 4a).

**Figure 4.** (**a**)Transient behavior of pure TiO2 nanotube array sensor and graphene-doped TiO2 nanotube array sensor in 100 ppm methanol at RT with measured response time and recovery time; Graphene-doped TiO2 nanotube array sensor; (**b**) Transient behavior from methanol concentration range of: −1000 ppm to 10 ppm; (**c**) Repeated cycles in 100 ppm methanol at RT.

A transient was measured within a concentration range of 1000 ppm to 10 ppm for a graphene-doped TiO2 nanotube array sensor (Figure 4b). A good response magnitude was obtained in 1000 ppm methanol (77.7%) and an average response magnitude was obtained in 10 ppm methanol (19%) at room temperature. The graphene-doped TiO2 nanotube array sensor exhibited a stable baseline resistance with highly repeatable transient behavior at room temperature (Figure 4c). The graphene-doped TiO2 nanotube array sensor response was improvised with short response time and recovery time due to the incorporation of graphene inside TiO2 nanotubes.

#### *3.3. Methanol Sensing Mechanism*

The large surface area and two dimensional structure of graphene enhanced the sensing performance of graphene-doped TiO2 nanotube array sensor. This increased conductance of graphene doped TiO2 nanotube array sensor can be attributed to the large carrier mobility and high electrical conductance of graphene. The uniform doping of graphene inside the TiO2 nanotubes improved the sensing parameters of graphene-doped TiO2 nanotube array and enabled room temperature sensing.

An energy band diagram of both junctions was sketched by considering the work function of GO *qϕGO* ∼ 4.5 eV [16] and anatase n-TiO2 *qϕTiO*<sup>2</sup> ∼ 5.1 eV [17]. An energy band gap of 3.59 eV for pure GO and 3.2 eV for pure TiO2 (S0) were estimated from a literature survey. On the formation of a heterojunction between TiO2 and GO, electrons are transferred to TiO2 and get accumulated on the TiO2 surface.

$$\text{O}\_2(\text{gas}) \rightarrow \text{O}\_2(\text{adsorbed}) \tag{1}$$

$$\text{O}\_2(\text{adsorbed}) + \text{e}^- \rightarrow \text{O}\_2^- \text{ (adsorbed)}\tag{2}$$

$$\rm O\_2^- + e^- \rightarrow 2O^- \text{(adsorbed)}\tag{3}$$

Surface adsorption of oxygen groups (O2 −, O−, O2−) reduces the electron concentration (Equations (1)–(3)) and increases the width of the depletion region, resulting in the formation of built-in potential on the surface of the graphene-doped TiO2 nanotube array sensor as represented in Figure 5. Upon exposure to the methanol vapors, the trapped electron oxygen groups are released back to the surface of the graphene-doped TiO2 nanotube array sensor, lowering the built-in potential.

$$\mathrm{CH\_3OH} + \mathrm{O^-} \rightarrow \mathrm{HCOOH} + \mathrm{H\_2O} + \mathrm{e^-} \tag{4}$$

$$\text{CH}\_3\text{OH} + \text{O}\_2^- \rightarrow \text{HCOOH} + \text{H}\_2\text{O} + \text{e}^- \tag{5}$$

When methanol vapor react with the oxygen species it gets oxidised into formaldehyde, then to formic acid and then it releases electrons to the conduction band which, in turn, reduces the resistance of the sensor in exposure to methanol vapors (Equations (4) and (5)) [18].

Formation of depletion region across the TiO2 and GO junction plays an important role for improving the sensor response. Uniform doping of graphene on the TiO2 surface is the main reason for enhancing the change of current in-between air and VOC ambient that eventually shows high sensitivity towards methanol by the graphene doped TiO2 nanotube sensor at room temperature with quick response time and recovery time.

**Figure 5.** Heterojunction formed between *p*-type GO and n-TiO2 nanotubes with electron depletion in GO and electron accumulation in TiO2.

#### **4. Conclusions**

In this work, electrochemical anodization was applied to develop pure TiO2 nanotube array and graphene-doped TiO2 nanotube array. Graphene was doped in the TiO2 nanotubes without hampering the original morphology of the nanotubes. Morphological characterization confirmed the formation of highly aligned and uniform nanotubes and structural characterization confirmed the anatase crystallinity of TiO2 nanotubes in both the samples. The evidence of graphene in the hybrid nanotubes was authenticated by the D and G peaks in the Raman spectra. The pure and graphene-doped TiO2 nanotube array sensor was produced in MIM structure where Au was considered as the top electrode and Ti was considered as the bottom electrode. Pure TiO2 nanotube array showed a response magnitude of 20% with slow response time (116 s) and recovery time (576 s) to 100 ppm methanol at room temperature. Graphene-doped TiO2 nanotube array showed a better response magnitude of 28% with a quick response time (34 s) and recovery time (40 s) to 100 ppm of methanol at room temperature. Also, lower detection limit till 10 ppm with good response magnitude (19%) towards methanol was achieved with the graphene-doped TiO2 nanotube array sensor at room temperature. A significant improvement in methanol sensing was achieved by the formation of localized heterojunctions between graphene and TiO2 in the hybrid sample.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/CSAC2021-10620/s1.

**Author Contributions:** Experiment, analysis and writing by T.G.; A.H. designed the research. Review editing and validation was also done by A.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported in part by Department of Biotechnology grant (Letter No. BT/PR28727/NNT/28/1569/ 2018), SPARC grant (SPARC/2018-2019/P1394/SL), and CSIR-SRF Direct (1431120/2k19/1) fellowship Govt. of India.

**Institutional Review Board Statement:** The study was conducted according to the guidelines and Deceleration of BITS Pilani.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** There are no conflicts to declare.

#### **References**


### *Proceeding Paper* **Morphological Effects in SnO2 Chemiresistors for Ethanol Detection: A Systematic Statistical Analysis of Results Published in the Last 5 Years †**

**Andrea Ponzoni**

National Institute of Optics of the National Research Council (CNR-INO), Unit of Brescia, 25123 Brescia, Italy; andrea.ponzoni@ino.cnr.it

† Presented at the 1st International Electronic Conference on Chemical Sensors and Analytical Chemistry, 1–15 July 2021; Available online: https://csac2021.sciforum.net/.

**Abstract:** SnO2 is one of the most studied materials in gas sensing. Among the many strategies adopted to optimize its sensing properties, the fine tuning of the morphology in nanoparticles, nanowires, and nanosheets, as well as their eventual hierarchical organization, has become an active field of research. In this work, results published in the literature over the last five years are systematically analyzed focusing on response intensities recorded with chemiresistors based on pure SnO2 for ethanol detection in dry air. Results indicate that no morphology clearly outperforms others, while a few individual sensors emerge as remarkable outliers with respect to the whole dataset.

**Keywords:** chemiresistors; SnO2; ethanol; nanoparticles; nanorods; nanosheets

#### **1. Introduction**

Chemiresistors based on semiconducting metal oxides are among the most popular gas sensing devices. Their success comes from their high sensitivity to a broad range of chemicals, their reduced size and power consumption, and their suitability for mass production at relatively reduced costs. To optimize the sensing layer, the fine control of the morphology, both at the level of individual nanostructures and at the level of their hierarchical assembly, has been reported as very effective [1,2].

In this work, with the aim to have a more general and reliable picture of the state of the art, results published in the literature in the last five years are systematically analyzed, focusing on response intensities recorded with chemiresistors based on pure SnO2 for ethanol detection in dry air, as the case example. In particular, we chose to focus on SnO2 because it is the most studied material among semiconducting metal oxides. Similarly, we chose ethanol as target gas because it is widely used as a test gas for the development of innovative materials (morphologies) and it is a key component in many applications [3].

#### **2. Materials and Methods**

This work considers the responses to ethanol reported for chemiresistors based on pure SnO2 in the period from January 2015 to July 2020. In order to have a common background between all the considered responses, only dry air tests have been taken into account.

The morphology of the SnO2 layer is described at two different levels: at the level of individual nanostructures and the level of their eventual hierarchical assembly.

Concerning the shape of individual crystallites composing the sensing layer, it has been categorized as follows:


**Citation:** Ponzoni, A. Morphological Effects in SnO2 Chemiresistors for Ethanol Detection: A Systematic Statistical Analysis of Results Published in the Last 5 Years. *Chem. Proc.* **2021**, *5*, 75. https://doi.org/ 10.3390/CSAC2021-10474

Academic Editor: Ye Zhou

Published: 30 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

• Nanosheets: thin nanostructures extending in two dimensions.

#### **3. Results**

As an example of the shape of elementary nanostructures widely investigated in the literature, Figure 1 reports the SEM images for two SnO2 layers composed by a disordered network of nanowires (Figure 1a), and by a disordered network of nanoparticles (Figure 1b) [1]. Therefore, some nanoparticles are distributed over the substrate individually, while others are distributed in μm-sized grains as a consequence of aggregation often observed in nanoparticle-based layers [1].

**Figure 1.** Examples of two different morphologies investigated in the literature for SnO2-based chemiresistors. (**a**) Film composed by a disordered network of SnO2 nanowires; (**b**) film composed by a disordered network of SnO2 nanoparticles, which are distributed either individually or in μm-sized aggregates. Reprinted from [1].

Boxplots resuming the responses to 10 ppm and to 300 ppm of ethanol reported in literature are shown in Figure 2a,b, respectively, grouping the results by nanostructure morphologies, namely nanorods, nanoparticles, and nanosheets.

**Figure 2.** Boxplots resuming the statistics of the response intensities of SnO2 chemiresistors grouped by crystallite shape. (**a**) Statistics recorded vs. 10 ppm of ethanol; (**b**) statistics recorded vs. 300 ppm of ethanol.

The statistical parameters describing these distributions are reported in Tables 1 and 2 for data shown in Figure 2a,b, respectively.


**Table 1.** Statistics of data shown in Figure 1a (responses to 10 ppm of ethanol).

**Table 2.** Statistics of data shown in Figure 1b (responses to 300 ppm of ethanol).


Statistical parameters reported in these tables are: the number of samples considered in each category (morphology of elementary nanostructures); the number of outliers identified for each category; the values of the 1st, 2nd, and 3rd quartiles (Q1, Q2, and Q3) of the response amplitude Ggas/Gair; and the values of the upper and lower whiskers. The pvalue of the median test comparing the median response of morphologies two by two are also reported in order to have a statistical check about the similarity and dissimilarity between median responses of the different morphologies.

#### **4. Discussion**

The distributions of the response intensities shown in Figure 2 depend on the gas concentration. This is partially due to the fact that different authors often tested their sensors against different ethanol concentration so there is no a complete overlap between concentration used in different articles. In other words, the sensors whose response is shown in Figure 2a are not exactly the same sensors whose response is shown in Figure 2b. Nonetheless, despite these differences, a common feature is that no morphology clearly performs better than other morphologies. Median tests reported in Tables 1 and 2 feature a p-value that is larger than 0.05 in all situations. This means that there is no clear evidence to reject the null hypothesis, i.e., there is no clear evidence to reject the hypothesis that the couple of morphologies under the test are not distinguishable. The same is observed for other concentrations and also considers the eventual hierarchical organization of the individual nanostructures into assemblies, such as hollow spheres, fibers, hollow fibers, etc. [46]. On the other hand, some materials emerge as outliers with respect to all morphologies. In Figure 2a, there are five outliers: four are the responses from layers composed by nanoparticles, namely [4–7] with response intensities of about 236, 50, 49, and 50 (to 10 ppm of ethanol), and one composed by nanosheets [33] featuring a response Ggas/Gair ≈ 50. As a reference, the median responses to this ethanol concentration are around 4.55, 2.3, and 10

for nanoparticles, nanorods, and nanosheets, respectively. Concerning the concentration of 300 ppm, four outliers emerges: the nanoparticles synthesized by [45], and two types of nanorods and the nanosheets developed by [38]. These materials feature responses of about 2000, 4070, 1609, and 495, compared with the median responses of 71, 52, and 38 for nanosheets, nanorods, and nanoparticles, respectively.

These results are arguably due to the longer tradition of the synthesis of nanoparticles with respect to those of nanowires and nanosheets. Such a longer experience may reasonably imply a more developed capability to effectively combine the many parameters underlying the sensing mechanism, which may counterbalance the advantages arising from the fine morphological tuning inherent in the more recent nanostructures.

**Funding:** His research was funded by Regione Lombardia and Fondazione Cariplo through the project EMPATIA@LECCO.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** All data are in the reported figures and in the cited references.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Proceeding Paper* **Review of the Recent Advances in Nano-Biosensors and Technologies for Healthcare Applications †**

**Maha Wajeeh Aqra <sup>1</sup> and Amall Ahmed Ramanathan 2,\***


**Abstract:** The growing human population and the discovery of new diseases and emerging pandemics have increased the need for healthcare treatments and medications with innovative designs. The emergence of nanotechnology provides a platform for novel diagnostic and therapeutic in vivo non-invasive detection and treatment of ailments. It is now the era of the Internet of things (IoT), and data acquisition and interpretation from various parts of the human body in real time is possible with interconnected sensors and information transfer devices. Miniaturization, low power consumption and price with compatibility to existing network circuits are essential requirements in the IoT. Biosensors made of nanostructured materials are the ideal choice due to the unique structural, chemical and electronic properties of these materials with the advantage of a large surface-to-volume ratio, which makes them very successful for use as sensors for the detection of diseases, drug carriers, filters, fillers and reaction catalysts in healthcare applications. In this paper, we reviewed the recent progress made in the research and applications of biosensors in health and preventive medicine. The focus of the paper is biosensors made of nanostructured layered materials such as graphene and its structural analogs molybdenum disulphide (MoS2) and boron nitride (BN). We discussed and highlighted the present capabilities of the different nano-forms of these materials in the detection and analysis of diseases. Their efficiencies in terms of the detection limit, the sensitivity and the adaptability to different environments were be discussed. In addition, the challenges and future perspectives of using nano-biosensors to develop efficient diagnostic, therapeutic and cost-effective monitoring devices with smart technologies were explored.

**Keywords:** electronic tongues and noses; 2D materials; nanopores; preventive medicine; non-invasive

#### **1. Introduction**

The detection of biological molecules, ions or species of interest (analyte) through the measurement and analysis of signals proportional to the concentration of the analyte is the basic function of a biosensor. The biological/chemical information needs to be transformed into readable outputs through the transducer. Biosensors used in the detection and prevention of diseases need to be non-invasive, highly selective, flexible and sensitive [1–3]. In addition in order to acquire and interpret signals from different parts of the body with interconnected or multifunctional sensors, the sensor design needs to be innovative and compatible with smart technologies that can transfer data with a high speed and accuracy [4,5]. Moreover, several constraints such as biocompatibility, reliability, stability, comfort, convenience, miniaturization and costs need to be considered [6]. The last decade has seen tremendous research on Two dimensional materials such as grapheme (Gr), graphene oxide (GrO) and molybdenum disulphide (MoS2) in different nano-forms for sensing applications in the healthcare, environment and other sectors [7–14].

Gr as the first 2D material discovered with its one-atomic-layer honeycomb structure has remarkable electronic, mechanical and optical properties and has seen a multitude of

**Citation:** Aqra, M.W.; Ramanathan, A.A. Review of the Recent Advances in Nano-Biosensors and Technologies for Healthcare Applications. *Chem. Proc.* **2022**, *5*, 76. https://doi.org/ 10.3390/CSAC2021-10473

Academic Editor: Xin Wu

Published: 30 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

applications [15–17]. Gr analogs MoS2 and boron nitride (BN) also have a honeycomb lattice and a layered structure that allow for the easy fabrication of 2D and other nanostructures due to the weak inter-layer van der Waal interactions.

A lot of research has been going on in the area of Gr and beyond Gr nanomaterials (NMs) during the past decade, and it is necessary to put into perspective and highlight the progress of Gr, MoS2 and BN nanostructures in biosensing for the healthcare sector. This is a rapidly changing and highly researched field with new discoveries and innovation and requires the frequent updates of progresses and challenges. This motivated us to present a focused review with the literature survey of the recent developments (last five years) on Gr and its structural analogs MoS2 and BN nanostructures in the detection and analysis of diseases in terms of efficiency, detection limits, sensitivity and adaptability to different environments. We discussed and highlighted the present capabilities of the different nanoforms of these materials. In addition, the challenges and future perspectives of using nano-biosensors to develop efficient diagnostic, therapeutic and cost-effective monitoring devices with smart technologies for healthcare and preventive medicine were explored. The article was arranged under the main headings: Introduction, Nano-Biosensors, Smart Technologies and Challenges or Opportunities.

#### **2. Nano-Biosensors**

#### *2.1. Biosensor Types*

Biosensors are two component devices consisting of a receptor and a transducer. The receptor is a biological recognition element which could be an enzyme, micro-organism, tissue, antibody or nucleic acid. The transducer converts the physiochemical change due to the interaction of the analyte with the receptor into an analytical output signal, which is coupled to an appropriate data-processing system. A schematic diagram of the process is shown in Figure 1.

**Figure 1.** Schematics of a biosensor unit.

Electrical, optical, electrochemical, micromechanical, calorimetric, magnetic, thermoelectric and piezoelectric transducers can be employed in biosensors, and the choice depends on the sensing environment and needs. Materials have been researched widely by the materials science community for use to fabricate the best-suited biosensor.

#### *2.2. Nanostructured Materials for Biosensing*

Gr and its analogs such as MoS2 and BN NMs have been the best materials for biosensing so far. The unique layered and honeycomb structure of these materials allows for the easy synthesis of monolayers (MLs), bilayers (BLs), nano-flakes, nanotubes and hetrostructures with a wide range of bandgaps and a diverse variety of optoelectronic properties. In addition, due to the weak inter-layer van der Waals forces, one can intercalate with atoms of different species and functionalize them easily to obtain the desired properties at will; moreover, NMs have the advantage of a large surface-to-volume ratio, which is important in the efficient immobilization of receptors on the surface of the NMs for good sensor performance [18]. All these factors make them prime candidates for use as biosensors in healthcare applications. In Figure 2, we give a graphical representation of NMs for biosensing that best describes the scope of this review.

**Figure 2.** Graphical representation of nanostructured materials for biosensing.

#### 2.2.1. Gr Nano-Biosensors

A Gr layer has a hexagonal symmetry with a honeycomb structure, and in-plane C atoms are bonded by strong covalent sp2 bonds with the nearest neighbours and an out-of-plane delocalised π bond, as shown in Figure 3a. It is the delocalised π electrons that are responsible for the extremely high room temperature mobility of 15,000–200,000 cm V−<sup>1</sup> s−<sup>1</sup> [19]. Moreover, Gr has an excellent mechanical strength on account of the strong covalent bonding and is optically transparent and highly flexible [20,21].

**Figure 3.** (**a**) Graphene geometry (**i**), bonding (**ii**) and the related band diagram (**iii**) [19]. (**b**) Schematic diagram showing the Dirac Fermi cone (**i**), the modification of the band by chemical or geometry restrictive doping (**ii**), the modification of the band by bilayer graphene (**iii**) and, finally, the modification of the bands in doped bilayer graphene (**iv**) [22].

The high electrical and thermal conductivity, mechanical strength, flexibility, optical transparency and ultrathin feature (one-atom thickness) of Gr are ideal characteristics for sensing applications. The sensor selectivity plays a very important role in its design, and this is very closely related to NM sensors characteristics, so selectivity can only be improved by fine-tuning the NM properties. The NM interacts with target bio-molecules by either a physisorption or chemisorption process. Physisorption, although fast, is a noncovalent bonding reaction and is not preferred as the bio-molecules do not bind completely, thereby affecting the sensitivity. Chemisorption can be brought about by the presence of defects, vacancies, doping and chemical functionalization, all of which increase the reactivity and enhance the selectivity to the target species. Figure 3b depicts the band structure changes of Gr by changing the geometry, thickness and doping mechanisms. The GrO 2D material produced by the oxidation of Gr is semiconducting and has a finite gap as compared to Gr. It has the advantage of being stable in water and other solvents and can be easily functionalized. Reduced graphene oxide (rGrO) is obtained by the removal

of the oxygen functional groups and has the advantages of Gr and GrO, which include being conducting and having chemically active defect sites. The bandgap engineering and chemical functionalizing of Gr through the use of Gr derivatives such as GrO and rGrO and composites have proved to work well as sensors (including wearable sensors and implantable devices) for human health monitoring, as reported in Table 1. The body temperature is an important indicator of abnormal body functions, and its measurement is one of the first lines of action in suspicious cases. We have seen ample evidence of this aspect during the current COVID-19 pandemic. It is also linked to our biological clock and can be used to monitor an individual's sleep patterns, which is important in determining the overall health and mental fitness. Table 1 gives a summary of various Gr-based sensors along with body functions tested, the mechanisms of sensing, the sensitivities and the ranges when available and the references.

**Table 1.** Summary of the details of graphene (Gr)-based sensors in health monitoring.


2.2.2. MoS2 and BN Nano-Biosensors

Similar to Gr, 2D MoS2 and hBN materials with a honeycomb structure have all the advantages of Gr for sensing mentioned in the previous section. These van der Waal structures exhibit unique optical and electronic properties that make them very appealing for biosensing [38]. Moreover, they have the added advantage of bandgaps unlike Gr which has a zero bandgap; this improves the sensitivity of sensor devices made of these materials, especially in sensors.

MoS2 is a prototype of a class of materials termed transition metal dichalcogenides (TMDs) and has markedly anisotropic properties, as seen from its electrical resistivity among other properties. The resistivity in a direction perpendicular to the planes is about 1000 times greater than in the parallel direction. Unlike Gr which is one-atom thick, an ML of MoS2 has three atomic layers sulfur–molybdenum–sulfur. The physical properties of MoS2 change markedly at the nanoscale. The bulk material has an indirect bandgap of ~1.2 eV, while the ML material has a direct and broader bandgap of ~1.8 eV [39]. Hence, it shows thickness-dependent bandgap properties, allowing for the production of tuneable optoelectronic devices with diversified spectral operation.

The electronic and optical properties of Gr and MoS2 are complemented by those of hBN, which is an insulator with a large indirect bandgap value of ~5.95 eV [40] in the bulk form and in the ML limit crossover to a direct-bandgap material with a gap of 6.1 eV [41]. The sensing mechanisms of these materials could be electrical-based sensing, through charge transfer which alters the resistance or optical sensing where due to the charge transfer the surface plasmon resonance (SPR) gets modified and can be detected; or biomolecules are detected by their spectral fingerprints.

The hybrid structures of Gr, MoS2 and BN have also been highly researched to increase the scope of the biosensing capabilities of these NMs. This is the topic of the next section.

#### 2.2.3. Hetrostructures

2D Gr, GrO, rGrO, MoS2 and hBN can all be used like Lego blocks to build interesting hetrostructures by mixing and matching for the increased selectivity and sensitivity of the nano-biosensensors. This process of electrostatic doping by the stacking of these van der Waal structures can be used to obtain unique and tuneable electronic properties. Figure 4 shows a graphical representation of a hetrostructure that can be made with the basic single layers of Gr, hBN and MoS2.

**Figure 4.** Graphical representation of possible hetrostructures that can be made by stacking multiple van der Waal layered structures in different orderings. Adapted from [42].

Hetrostructures, although highly desirable, require careful considerations of the lattice mismatch, the misalignment of layers and the introduction of unforeseen defects during the deposition and the epitaxial growth. Hexagonal BN is an insulating analogue of graphite with a small lattice mismatch (~1.8%), so it is an ideal substrate for graphene and a key building block in many van der Waals hetrostructures. Gr–hBN-integrated devices have been recently used for DNA sequencing by current modulation [43] and distinguishing nucleotides in DNA [44]. An SPR-based biosensor consisting of Gr/hBN hybrid structures for the detection of biomolecules was reported in 2019 [45]. The SPR technique is also used in a biosensor consisting of a MoS2/Gr hybrid structure with Au, as a substrate, used to detect biomolecules using SPR [46]. Again in 2017, an angle-based SPR biosensor made of a MoS2/Al film/MoS2/Gr heterostructure was used to detect biomolecules [47]. In Table 2, we summarized various nano-biosensors made of MoS2, hBN, Gr, Gr derivatives and hetrostructures of these NMs in the recent years. Table 2 gives the NMs used, the species detected, the sensing mechanisms, the sensitivities, the detection ranges, publications and the years of publications.


**Table 2.** Summary of nano-biosensors with references and the years of research.

#### **3. Smart Technologies**

The early-stage detection and prevention of chronic and fatal diseases requires continuous monitoring. Data acquisition and interpretation from various parts of the human body in real time is possible with interconnected sensors and information transfer devices in today's era of the Internet of things (IoT). The unprecedented advancements in electronics and sensor technologies coupled with Big Data and AI offer exciting opportunities in the field of smart and sustainable healthcare. The stage is now set to shift from old medical procedures and protocols and adapt smart integrated medical testing with nano-devices for diagnosis and therapeutics [61,62]. We need to discard costly and bulky equipment and old fashioned laboratories and embrace wearable and miniaturised sensors that use interstitial fluid (ISF), instead of blood, to detect minute changes in biomarkers with sweat, tears and breath analysis that contain a wealth of information about body malfunctions [1–3]. Wireless, powerless nano-devices made of biocompatible materials that can be worn on the skin (patches, tattoos, watches, etc.), in textiles, the eye, mouth, teeth (miniaturised implants) and other innovative means using non-invasive probes are the need of the day. Electronic nose, tongues and skin are the new innovative smart technologies that are the future of healthcare monitoring and preventive medicine [63–67].

#### **4. Challenges or Opportunities**

A challenge, limitation or drawback is an opportunity for improvement, change in strategy or chance for innovation. Although the nano-biosensors research shows that considerable improvements in healthcare monitoring can be made, commercial products are few and from small companies [68]. Before large-scale and widespread manufacturing of 2D and other nanostructured devices for health-related applications can be realized, uniformity and controlled synthesis is necessary to rule out the device-to-device variability. This is crucial for large-scale commercialization, and the challenge has been met as indicated by the recent research and publications addressing this issue [69,70].

In addition, in vivo and point-of-care diagnostics require biocompatibility and toxicity issues to be addressed. The precise control of NMs properties and biocompatibility is required, especially in the local biological environment, where the devices are to be used with a thorough understanding of complex physiochemical interactions. The recent years have seen tremendous work in this direction with good progresses [71–73].

**Author Contributions:** A.A.R. conceived, designed and wrote the project manuscript; M.W.A. contributed towards literature survey, curation of data and writing. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors have no conflict of interests to declare.

#### **References**


### *Proceeding Paper* **Feasibility of Total White Blood Cells Counts by Visible-Near Infrared Spectroscopy †**

**Teresa Guerra Barroso <sup>1</sup> , Lénio Ribeiro 2,3, Hugo Gregório <sup>2</sup> , Filipe Santos <sup>4</sup> and Rui Costa Martins 4,\***


**Abstract:** Total white blood cells (WBC) count is an important indication for infection diagnosis, in both human and veterinary medicine. State-of-the-art WBC counts are performed by flow cytometry combined with light scattering or impedance measurements, in the clinical analysis laboratory. These technologies are complex and difficult to be miniaturized into a portable point-of-care (POC) system. Spectroscopy is one of the most powerful technologies for POC miniaturization due to its capacity to analyze low sample quantities, little to no sample preparation, and 'real-time' results. WBC is in the proportion of 1:1000 to red blood cells (RBC), and the latter dominate visible-near infrared (Vis-NIR) information due to their large quantities and hemoglobin absorbance. WBC are difficult to be detected by traditional spectral analysis because their information is contained within the interference of hemoglobin bands. Herein, we perform a feasibility study for the direct detection of WBC counts in canine blood by Vis-NIR spectroscopy for veterinary applications, benchmarking current chemometrics techniques with self-learning artificial intelligence—a new advanced method for high-accuracy quantification from spectral information. Results show that total WBC counts can be detected by Vis-NIR spectroscopy to an average detection limit of 7.8 <sup>×</sup> <sup>10</sup><sup>9</sup> cells/L, with an R<sup>2</sup> of 0.9880 between impedance flow cytometry analysis and spectral quantification. This result opens new possibilities for reagent-less POC technology in infection diagnosis. As WBC counts in dogs range from 5 to 45 <sup>×</sup> <sup>10</sup><sup>9</sup> cells/L, the detection limit obtained in this research allows concluding that the combined use of spectroscopy with this SL-AI new algorithm is a step towards the existence of portable and miniaturized Spectral POC hemogram analysis.

**Keywords:** point-of-care; spectroscopy; white blood cells; artificial intelligence

#### **1. Introduction**

Total white blood cell (WBC) count is one of the most requested hematology parameters because of its broad diagnostic value, including for infection and leukemia. Leukocytosis and leukopenia, which are abnormal values (high/low, respectively) in WBC counts, are more frequently associated with neutrophil changes, although other leukocytes and neoplastic cells can also cause fluctuations. Neutrophilia is usually related to inflammation, and neutropenia to greater peripheral use or reduced bone marrow production [1].

Most common methods for WBC differential are based on electrical impedance, laser light scattering, radio frequency conductivity, and/or flow cytometry [2] (Figure 1). The basic principles of operation for automated hematology analyzers are based on cell size

**Citation:** Barroso, T.G.; Ribeiro, L.; Gregório, H.; Santos, F.; Martins, R.C. Feasibility of Total White Blood Cells Counts by Visible-Near Infrared Spectroscopy. *Chem. Proc.* **2021**, *5*, 77. https://doi.org/10.3390/ CSAC2021-10434

Published: 30 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

affecting directly impedance and scattering angle. This approach has disadvantages for WBC differential, because cell sizes for each type of leukocyte are highly dependent on the development stage and differentiation, leading to inaccurate counts in current automated equipment [3]. Despite laser scattering technology provides better accuracy than impedance technology, the latter is widely adopted in Veterinary Medicine. Impedance counting is a cheaper technology and the best hematology practices recommend that blood smear microscope counts are performed on abnormal cases [4].

Spectroscopy is one of the leading technologies for the development of reagentless point-of-care (POC) devices [5,6], capable of providing comprehensive clinical information from a single drop of blood (<10 μL), with little or no sample preparation and real-time results.

Visible short-wave near-infrared (Vis-SWNIR) spectroscopy is an information-rich technology that carries both physical and chemical information, where the information about blood cells and constituents is distributed across the different wavelengths. Dominant spectral information in blood comes from highly absorbent constituents in the Vis-SWNIR region, such as hemoglobin present in red blood cells (RBC) and bilirubin in serum.

WBC is present in significantly lower quantities than RBC (∼1:1000), being considerably more difficult to be detected because the information about WBC is a small interference effect on the hemoglobin bands. State-of-the-art chemometrics and artificial intelligence technologies are unable to deal with small-scale interference and non-dominant spectral information sample constituents with good accuracy [6]. Such may lead to non-causal correlation in spectroscopy quantification, where the quantification is not obtained by direct relationship to the spectral absorbance bands, but rather by intrinsic correlations of the dataset [7], which may lead to erroneous diagnosis [6].

**Figure 1.** Total white blood cell counts: (**a**) current laboratory methods—automated cell counting using electric impedance or laser scattering, and manual smear count at the microscope by trained hematologist; and (**b**) Point-of-care approach—single blood drop spectroscopy counts using artificial intelligence.

In this research, we study the capacity of WBC quantification by Vis-SWNIR spectroscopy and a new algorithm based on Self-Learning Artificial Intelligence [6]. This new approach isolates spectral interference by searching consistent covariance between WBC and spectral features—the covariance mode (CovM). CovM is a set of samples that allow the direct relationship between spectral features and WBC, by sharing the same latent structure information [6]. Ideally, the relationship between WBC and spectral features

is given by a single eigenvector or latent variable (LV), allowing to unscramble spectral interference in complex samples such as blood.

Herein, we provide a feasibility study on using Vis-SWNIR spectroscopy for the quantification and diagnosis of WBC, by providing a benchmark between a common chemometrics technique—partial least squares (PLS), and our new methodology (SL-AI).

#### **2. Materials and Methods**

#### *2.1. Hemogram Analysis*

Blood samples from daily clinical practice were collected from the jugular vein by qualified personnel using standardized venipuncture procedures at the Centro Hospitalar Veterinário do Porto into EDTA tubes. The sample was measured to WBC by a Beckman– Coulter capillary impedance [8] Mindray BC-2800-vet auto-hematology analyzer (Mindray, Shenzen, China), and a drop of blood (10 μL) was used for spectroscopy measurement.

#### *2.2. Spectroscopy*

Blood spectra were recorded using a POC prototype (INESC TEC, Porto, Portugal) using a 4500 K power LED as light source, and an USB-based miniaturized spectrometer (Ocean Insight STS-vis, Orlando, FL, USA) with an optical configuration and plug-in capsule system according to [5]. LED temperature and spectrometer integration times were automatically managed to maintain result consistency. Three replicates measurements were made for each blood sample.

#### *2.3. Chemometrics*

Spectral records were subjected to scattering correction (Mie and Rayleigh) before modeling. A feasibility benchmark is performed between PLS and SL-AI methods. PLS maximizes the global covariance between spectral features and WBC, by determining the orthogonal eigenvectors of the covariance matrix. The relationship between WBC and signal features is derived by the latent variables (LV), at each deflation. The number of LV is determined by cross-validation at the minimum value of the predicted residuals sum of squares (PRESS) [9].

SL-AI searches for stable covariance in spectral datasets, finding covariance modes (CovM). CovM is a group of samples that hold the same interference information characteristics, carrying proportionality between WBC and spectral features. Ideally, the CovM relationship between WBC and spectral features is given by a single eigenvector or latent variable (LV). The CovM is validated by leave one-out cross-validation [6].

#### **3. Results and Discussion**

PLS attains a correlation of 0.5687 and a SE of 11.60 × <sup>10</sup><sup>9</sup> cell/L (Table 1). PLS analysis demonstrates that there is a significant correlation between spectral features and WBC, and the small-scale interference of WBC is present in the spectra records. PLS model is obtained with 5 LV. Such means that the interference information about WBC in the blood Vis-SWNIR spectra is present in a significant number of differentiated covariance modes, where the non-dominant spectral interference can be related to WBC. PLS collapses the 5 LV into a single linear coefficient, which relates the WBC to the recorded spectra, leading to an averaged representation of all covariance modes present in the dataset. Such results in a high SE and MAPE of 44.62%. The PLS model is unable to estimate WBC values above 45.00 × 109 cell/L, misdiagnosing severe infection cases (Figure 2).

**Figure 2.** Total white blood cell counts spectral quantification: (**a**) PLS and (**b**) SL-AI.

The minimal total error criteria established by the American Society for Veterinary Clinical Pathology (ASVCP) for WBC is 20%. PLS shows to be unable to provide the necessary accuracy for WBC spectral POC technology.

SL-AI has a significantly higher correlation (R = 0.9733), a SE of 2.16 × <sup>10</sup><sup>9</sup> cell/L, and a MAPE of 20.00%. SL-AI covariance modes are obtained with 3 LV (Table 1). Results show that the different covariance modes (CovM) hold spectral interference proportional to WBC. Such demonstrates that it is possible to search non-dominant spectral interference from WBC and correlate it to total WBC count (Figure 3).

SL-AI CovM relationships are obtained with 3 LV. This is an indication that interference with other constituents and WBC differential population are incorporated in total WBC count, and that this higher complexity is not completely unscrambled in the dataset. In ideal conditions, CovM is obtained with a single LV (one eigenvector), directly relating the constituent concentration to spectral interference. The results show that non-dominant WBC spectral interference information has high complexity, which can be attributed to complex immune response, where differentiated cell types act at different stages and levels of infection or inflammation. The LV number re-assures the need for further studies, in order to investigate the source of non-dominant spectral interference attributed to WBC. Results may be improved by:


Despite the limitations shown in this feasibility study, WBC quantification using Vis-SWNIR spectroscopy in conjunction with the new SL-AI algorithm can attain a total error estimate of 20%. Such result is following the ASVCP total allowable error for WBC in dog blood [4], but is above the 15% total allowable error in humans defined by CLIA [10].

**Table 1.** WBC quantification benchmarks for PLS and SL-AI.


**Figure 3.** Percentage total error for PLS and SL-AI predictions: (1) ASVCP acceptable error limit (20%) and (2) CLIA acceptable error limit (15%).

#### **4. Conclusions**

This feasibility study has shown that low intensity, non-dominant, and multi-scale interferent spectral information is possible to be accessed by unscrambling information with the CovM principle included in our SL-AI method. The smaller quantities of WBC and corresponding interference with dominant constituents, such as erythrocytes, hemoglobin, and bilirubin, are detectable in each CovM. The results allow us to conclude that a spectral POC in the Vis-SWNIR for measuring WBC is achievable, for the application in both veterinary and human medicine.

**Author Contributions:** T.G.B., L.R. and H.G.: Investigation, methodology, validation, writing review and editing; F.S.: investigation, hardware and firmware; R.C.M.: conceptualization, software and hardware, funding acquisition, writing—original draft, resources and formal analysis, project administration. All authors have read and agreed to the published version of the manuscript.

**Funding:** Rui Costa Martins acknowledges Fundação para a Ciência e Tecnologia (FCT) research contract grant (CEEIND/017801/2018).

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

#### **References**


### *Proceeding Paper* **Visible–Near-Infrared Platelets Count: Towards Thrombocytosis Point-of-Care Diagnosis †**

**Teresa Guerra Barroso <sup>1</sup> , Lénio Ribeiro 2,3, Hugo Gregório <sup>2</sup> , Filipe Santos <sup>4</sup> and Rui Costa Martins 4,\***


**Abstract:** Thrombocytosis is a disorder with an excessive number of platelets in the blood, where total platelet counts (TPC) are crucial for diagnosis. This condition predisposes to blood vessels clotting and diseases such as stroke or heart attack. TPC is generally performed at the laboratory by flow cytometry with laser scattering or impedance detection. Due to the limited capacity of automated hematology in performing TPC quantification, a manual microscopy count is a very common quality assurance measure undertaken by clinical pathologists. Monitoring coagulation risk is key in many health conditions, and point-of-care platforms would simplify this procedure by taking platelet counts to the bedside. Spectroscopy has high potential for reagent-less point-of-care miniaturized technologies. However, platelets are difficult to detect in blood by standard spectroscopy analysis, due to their small size, low number when compared to red blood cells, and low spectral contrast to hemoglobin. In this exploratory research, we show that it is possible to perform TPC by advanced spectroscopy analysis, using a new processing methodology based on self-learning artificial intelligence. The results show that TPC can be measured by visible–near-infrared spectroscopy above the standard error limit of 61.19 <sup>×</sup> 109 cells/L (R<sup>2</sup> = 0.7016), tested within the data range of 53 <sup>×</sup> 109 to 860 <sup>×</sup> <sup>10</sup><sup>9</sup> cells/L of dog blood. These results open the possibility for using spectroscopy as a diagnostic technology for the detection of high levels of platelets directly in whole blood, towards the rapid diagnosis of thrombocytosis and stroke prevention.

**Keywords:** point-of-care; spectroscopy; platelets; artificial intelligence

#### **1. Introduction**

Platelets (PLT) are the smallest cells in the blood, being responsible for coagulation and blood vessel repair. The PLT counts reference interval in dogs is 300 to 500 × 109 cell/L. High PLT counts is a condition known as thrombocytosis, being attributed to abnormal bone marrow production or an ongoing condition such as anemia or inflammation [1]. Thrombocytosis can result in blood clots, leading to life-threatening or impairing conditions such as heart attack or stroke [2]. Automated PLT counts are mostly performed by flow cytometry, electric impedance (Coulter principle), or laser-scattering technologies [3]. However, these methods are prone to erroneous PLT counts, because of changes in cell size and morphology, due to blood clotting, activation, aggregation, or even post-sampling artifacts. This limits scattering angle and impedance detection, leading to misidentification

**Citation:** Barroso, T.G.; Ribeiro, L.; Gregório, H.; Santos, F.; Martins, R.C. Visible–Near-Infrared Platelets Count: Towards Thrombocytosis Point-of-Care Diagnosis. *Chem. Proc.* **2021**, *5*, 78. https://doi.org/ 10.3390/CSAC2021-10435

Academic Editor: Manel del Valle

Published: 30 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

as larger cells, such as erythrocytes or leucocytes. Laser scattering is significantly more accurate than electric impedance, but the latter is cheaper and has a higher implementation in veterinary medicine. Veterinary doctors make use of blood-smear PLT manual counts for ensuring results quality in abnormal (low or high) values [4].

Visible-shortwave–near-infrared (Vis–NIR) spectroscopy has a high potential for the development of point-of-care (POC) without the need for reagents or complex sample preparation. The developed Vis–SWNIR POC system (Figure 1b) records the blood spectra of a single drop of blood (<10 μL) to provide a significant number of clinical analysis parameters with real-time results [5].

Visible-short-wave–near-infrared (Vis–SWNIR) spectroscopy is an information-rich technology that carries both physical and chemical information, where the information about blood cells and constituents is distributed across the different wavelengths. Dominant spectral information in blood comes from highly absorbent constituents in the Vis–SWNIR region, such as hemoglobin present in red blood cells (RBC) and bilirubin in blood serum.

Platelets are present in significantly lower values than red blood cells (RBC) (Figure 1a). The PLT reference interval in dogs is 300 to 500 × 109 cells/L and RBC is 5500 to 8500 × 109 cells/L, being at approximately 1:18 ratio to RBC, which makes the detection difficult:


**Figure 1.** Platelets cell counts: (**a**) manual smear count at the microscope by trained hematologist demonstrating the proportionality between (1) platelets, (2) white blood cells, and (3) red blood cells and (**b**) point-of-care approach—single-blood-drop spectroscopy counts using artificial intelligence.

PLT counts are difficult to obtain, even by microscopy methods, exhibiting high variability. Herein, we explore the capacity of Vis–SWNIR and self-learning artificial intelligence (SL-AI) for PLT quantification [5]. This new approach isolates spectral interference by searching consistent covariance between PLT and spectral features, which belong to a covariance mode (CovM). CovM is a set of samples that can hold a direct relationship between spectral features and PLT counts, by sharing a common latent structure [5]. Ideally, PLT counts are related to spectral-interference features by a single latent variable (LV) or eigenvector. This allows unscrambling the interference of PLT concerning the other blood

constituents. This research provides a feasibility benchmark between the widely used chemometrics partial least squares (PLS) method and the SL-AI method.

#### **2. Materials and Methods**

#### *2.1. Hemogram Analysis*

Dog blood samples from routine clinical practice were collected by qualified personnel by standard venipuncture, at the Centro Hopitalar Veterinário do Porto. No animal experimentation was involved or any additional procedure. Samples used in this study are remnants from already necessary routine clinical analysis medical practice. Dataset is anonymized. PLT was determined by Beckman–Coulter capillary impedance using a Mindray B-2800 vet auto-hematology analyzer (Mindray, Shenzen, China).

#### *2.2. Spectroscopy*

Blood spectra were recorded using a POC prototype (INESC TEC, Porto, Portugal) using a 4500 K power LED as a light source and a USB-based miniaturized spectrometer (Ocean Insight STS-vis, Orlando, FL, USA), with an optical configuration and plug-in capsule system according to [6]. LED temperature and spectrometer integration times were automatically managed to maintain result consistency. Three replicate measurements were made for each blood sample.

#### *2.3. Chemometrics*

Spectral records were subjected to scattering correction (Mie and Rayleigh) before modeling. A feasibility benchmark was performed between PLS and SL-AI methods. PLS maximizes the global covariance between spectral features and PLT, by determining the orthogonal eigenvectors of the covariance matrix. The relationship between PLT and signal features is derived by the latent variables (LV), at each deflation. The number of LV is determined by cross-validation at the minimum value of the predicted residuals sum of squares (PRESS) [7].

SL-AI searches for stable covariance in spectral datasets, finding covariance modes (CovM). CovM is a group of samples that contains the same interference information characteristics, holding proportionality between PLT and spectral features. Ideally, the relationship between PLT and spectral features is given by a single eigenvector or latent variable (LV). The CovM is validated by leave-one-out cross-validation [5].

#### **3. Results and Discussion**

The PLS model attained a correlation of 0.2613 with a very poor R2 (0.068) and a corresponding high SE of 175.99 × 109 cells/L. The PLS analysis shows that the correlation between spectral features and PLT counts is highly unstable and non-linear. Such is because PLT is present in much fewer quantities than other blood constituents (Figure 1), as well as, due to the small size and high interference with the other major blood constituents (e.g., RBC, hemoglobin, and bilirubin). Another indication of non-linearity is that the PLS algorithm attains the optimum prediction error with two LV, resulting in a non-significant model (Figure 2a). The PLS is unable to increase the number of LV because the information about PLT is scattered in significantly different interference modes that cannot be collapsed into a linear oblique projection model [5,7]. PLS cannot be used in a POC as it does not attain a MAPE similar to 25%—the total allowable error established by the American Society for Veterinary Clinical Pathology (ASVCP) for PLT counts [8].

SL-AI presented a significant correlation of 0.8376, an SE of 61.19 × 109 cell/L, and a MAPE of 24.67%, with R<sup>2</sup> of 0.7016 (Table 1). SL-AI covariance modes (CovM) were obtained with 1 to 3 LV. This means that, although statistically valid relationships were obtained for each CovM, some of these were integrating more than one type of interference. Under ideal conditions, all CovM should have only one LV, directly relating PLT counts and spectral interference.


**Table 1.** PLS and SL-AI benchmark results.

**Figure 2.** Total platelet counts spectral quantification: (**a**) PLS and (**b**) SL-AI.

**Figure 3.** Percentage total error for PLS and SL-AI predictions: (1) ASVCP acceptable error limit (25%).

The results also show that non-dominant spectral information and low-scale spectral variation is unscrambled by the CovM principle. The number of LV can be attributed to the high diversity of PLT morphology present in dog blood (non-activated, activated, and clotted PLT) and the particular conditions of the tested blood, with correspondence in the major constituents.

Despite the limitations shown in this feasibility study, PLT quantification using Vis– SWNIR spectroscopy in conjunction with the new SL-AI algorithm can attain a total error estimate of 25%. Such a result is following the ASVCP total allowable error for PLT in dog blood [8] (Figure 3).

Vis–SWNIR POC technology based on SL-AI has shown high potential for PLT quantification and thrombocytosis diagnosis. The results presented for dog blood are within the acceptable error defined by the ASVCP of 25% [8]. The presented results also allow extending the potential application to both human and other animal species in further studies.

#### **4. Conclusions**

This feasibility study showed that low intensity, non-dominant, and multi-scale interferent spectral information is possible to be accessed, by unscrambling information with the CovM principle included in the SL-AI method. The small variations in the spectral signal

that contain information about PLT cannot be modeled by PLS. SL-AI can unscramble PLT interference information based on the CovM principle, allowing the quantification of PLT. Future studies, with more samples, may provide better insights on the full potential of the developed POC technology in both veterinary and human medicine.

**Author Contributions:** T.G.B., L.R. and H.G.: investigation, methodology, validation, writing—review and editing; F.S.: investigation, hardware, and firmware; R.C.M.: conceptualization, software and hardware, funding acquisition, writing—original draft, resources and formal analysis, and project administration. All authors have read and agreed to the published version of the manuscript.

**Funding:** Rui Costa Martins acknowledges Fundação para a Ciência e Tecnologia (FCT) research contract grant (CEEIND/017801/2018).

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

#### **References**


### *Proceeding Paper* **Development of an Integrated In-Vehicle Driver Breath Ethanol System Based on** α**-Fe2O3 Sensing Material †**

**Roberto Di Chio \*, Monica Galtieri, Nicola Donato and Giovanni Neri**

Department of Engineering, University of Messina, Contrada Di Dio, 98166 Messina, Italy; monicagaltieri94@gmail.com (M.G.); nicola.donato@unime.it (N.D.); giovanni.neri@unime.it (G.N.)

**\*** Correspondence: rdichio@unime.it

† Presented at the 1st International Electronic Conference on Chemical Sensors and Analytical Chemistry, 1–15 July 2021; Available online: https://csac2021.sciforum.net/.

**Abstract:** Alcohol abuse is the dominant cause of fatal car accidents (about 25% of all road deaths in Europe). The large-scale implementation of systems aimed at the realization of in-vehicle driver breath ethanol detection is therefore in high demand. For this reason, we devoted our attention to the design of an inexpensive and reliable breath alcohol sensor for use in an Advanced Driver Assistance System (ADAS). The main challenge in the development of this sensor is related to the complexity of breath composition and its high humidity content, coupled with the high dilution of breath reaching the sensor. In this work, a simple α-Fe2O3 film-based sensor was developed and validated in laboratory tests. Tests were also performed by placing the ethanol sensor within the casing of the upper steering column of a car to simulate real driving conditions. Using an array provided with the developed ethanol sensor and humidity, temperature and CO2 sensors, it was possible to differentiate the signal of a driver's breath before and after alcohol consumption.

**Keywords:** gas-sensing; ethanol; iron oxide; sensing materials; ADAS

**Citation:** Di Chio, R.; Galtieri, M.; Donato, N.; Neri, G. Development of an Integrated In-Vehicle Driver Breath Ethanol System Based on α-Fe2O3 Sensing Material. *Chem. Proc.* **2021**, *5*, 79. https://doi.org/ 10.3390/CSAC2021-10476

Academic Editor: Huangxian Ju

Published: 30 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

#### **1. Introduction**

Advanced Driver Assistance Systems (ADASs) are intelligent systems that assist the driver in a variety of ways [1]. They may be used to provide useful traffic information but may also be used to evaluate whether or not the driver is in physical condition to drive. Among other driver-related risk factors (e.g., drug intake or altered emotional state), alcohol abuse remains the dominant cause of fatal car accidents (about 25% of all road deaths in Europe). It is well known that too much alcohol in the blood leads to various serious effects on human health and the condition of drivers [2,3]. Alcohol interferes with the brain, affecting the way that it looks and works and reducing movement coordination. Further, alcohol can slow reflexes, slow eye muscle function, and alter visual perception. These conditions are very critical for car drivers, so maintaining an acceptable blood alcohol level is necessary in order to limit car accident risks.

Based on these concerns, we initiated research activity with the main objective of developing an in-vehicle driver breath ethanol detection system [4]. To facilitate the large-scale implementation of these systems, the design of inexpensive, reliable and easily fabricated sensors is required. Conductometric sensors apply very well to this scope, as they possess all of the required characteristics [5]. Many examples of ethanol sensors have been developed and show a remarkable sensing capacity [6–8]. In particular, we have shown that α-Fe2O3 is an ideal candidate as a sensing material to be used in breath ethanol conductometric sensors [9,10].

Based on previous work, in this research, the α-Fe2O3 material was employed for fabricating conductometric gas sensors to be used for breath ethanol detection in ADASs. Preliminary laboratory tests were performed to validate the fabricated sensors and optimize

the operating conditions. Then, tests were performed by placing the ethanol sensor within the casing of the upper steering column of a car to simulate the driving position. The main challenge in the development of this system is related to the complexity of breath composition and its high humidity content, coupled with the high dilution of breath reaching the sensor. For this reason, it was necessary to install the ethanol sensor in an array that also contains humidity, temperature and CO2 sensors (the latter breath component is employed as an internal standard). Through the simultaneous use of these three sensors, it was possible to differentiate the signal of a driver's breath before and after alcohol beverage consumption.

#### **2. Materials and Methods**

#### *2.1. Material Preparation*

For the synthesis of α-Fe2O3 material, a simple Pechini sol–gel process was employed [6,7]. This method is based on the polymerization of metallic citrate by ethylene glycol. Iron nitrate (Fe(NO3)3·9H2O), citric acid (C6H8O7 H2O), poly(vinylpyrrolidone) and ethylene glycol (C2H6O2) were purchased from Merck. All of the chemicals were used as received and without further purification. Double-distilled water was used to prepare precursor solutions.

First, the appropriate amount of Fe(NO3)3·9H2O was dissolved in distilled water at 70 ◦C for 1 h under magnetic stirring to produce a 0.5 M Fe+3 solution. Then, this solution was mixed with PVP solution with a molar ratio of [PVP]/[Fe+3] = 1. On the other hand, citric acid was dissolved in distilled water at 70 ◦C for 30 min. Afterwards, the citric acid solution was added slowly to the Fe+3/PVP solution with stirring. The citric acid to Fe+3 molar ratio was 2. Then the esterification agent, i.e., ethylene glycol (EG), was added with a molar ratio of [citric acid]/[EG] = 2 while stirring and heating the solution. The final solution was refluxed at 100 ◦C for 2h. The clear yellow-colored precursor solution obtained was dried at 120 ◦C for 12 h to obtain the precursor powders. Finally, the amorphous powders were calcined at 550 ◦C in air for 3 h using a muffle furnace to obtain iron oxide nanoparticles.

#### *2.2. Sensor Preparation and Sensing Tests*

Sensor devices were fabricated by the spray-coating method as follows. An appropriate volume of the α-Fe2O3 suspension was sprayed on alumina substrates (3 × 6 mm) supplied with interdigitated Pt electrodes and a heating element on the backside. The prepared sensors were dried at room temperature and then heat-treated at 400 ◦C to obtain a mechanically stable sensing layer. The structure of the fabricated ethanol sensor is shown in Figure 1.

**Figure 1.** Photo of the fabricated ethanol sensor.

Measurements were performed under both a dry and wet (50% relative humidity) air stream of 100 mL/min in total, and the sensor resistance data were collected in fourpoint mode using an Agilent 34970A multimeter. Electrical measurements were carried out at a working temperature of 300 ◦C. Laboratory sensing tests were performed in a lab apparatus that allows operation at controlled temperature and the performance of resistance measurements while varying the ethanol concentration from 12.5 to 400 ppm.

The gas response was defined as the ratio Rair/Rgas, where Rair represents the electrical resistance of the sensor in dry air, and Rgas is the electrical resistance of the sensor at different ethanol concentrations. Response time, tres, was defined as the time required for the sensor resistance to reach 90% of the equilibrium value after ethanol is injected, and recovery time, trec, was taken as the time necessary for the sensor resistance to reach 90% of the baseline value in air.

#### **3. Results**

#### *3.1. Laboratory Sensing Tests*

The characteristics of the developed α-Fe2O3 sensor were first evaluated in laboratory tests. Based on the preliminary results, a temperature of 300 ◦C was selected as the operating temperature. Figure 2a shows the sensor behavior versus ethanol concentration, which ranged from 400 to 12.5 ppm at this temperature. A reversible variation in the resistance was observed with the concentration of ethanol. As usually verified for metal oxide-based conductometric sensors, response and recovery times are dependent on the alcohol concentration. This is also the case for our sensor. Response (8–15 s) and recovery (120–45 s) times were observed for ethanol concentrations ranging from 400 to 12.5 ppm. At the intermediate concentration of 100 ppm, the sensor showed a noticeable reversible response (see Figure 2b) with a fast response and recovery (about 10 s and 60 s, respectively).

**Figure 2.** (**a**) Response of the sensor to a variable concentration of ethanol in dry air at 300 ◦C; (**b**) response of the sensor to an ethanol pulse of 100 ppm. The measured response and recovery times are reported.

From the above test, the calibration curve shown in Figure 3 was obtained. Plotting the data in a log–log graph, a high linear correlation between the sensor resistance and the ethanol concentration is observed. The same graph also shows the calibration curve for the same sensor obtained in conditions of higher relative humidity (50% RH). Breath is highly saturated with water vapor; therefore, the sensor performance must not be influenced by changes in the humidity level [11–13]. Interestingly, the sensor signal that we collected in different humidity conditions appears to be independent of this variable.

**Figure 3.** Calibration curve at 300 ◦C for the α-Fe2O3 sensor in *dry* and *wet* conditions.

#### *3.2. Ethanol Sensor Implementation in ADASs*

Then, the research work continued with the installation of the ethanol sensor in the casing of the upper steering column of a car to simulate real driving conditions (see Figure 4). A diagram of the designed and constructed module consisting of the ethanol sensor used in this research was reported in a previous paper [14]. Humidity, temperature and CO2 sensors were also installed. The detected CO2 concentration was used to account for the dilution of the breath sample. A suitable chamber was therefore designed and built to contain the sensor array.

**Figure 4.** Pictures showing the location of the ethanol sensor inside the casing of an upper steering column (**left**) and the position of the driver during the test, simulating real driving conditions (**right**).

After installing the sensors, some preliminary tests to validate their correct functioning were carried out, especially to verify if the breath of the driver can be well detected by the sensor array when it is located at a distance of 30–50 cm from the driver's mouth. Indeed, in the conditions adopted, breath is diluted with ambient air by a factor as high as 5–10 [15,16].

One subject (male, 70 kg) was used for the test. He was allowed to drink two glasses (50 mL in total) of a commercial alcoholic beverage (44% of ethanol in *v*/*v*) in less than 15 min, for a total weight of 17.5 g of ingested alcohol (0.25 g of alcohol per kilogram of body mass). Measurements were performed before drinking and repeated every 30 min until the breath alcohol level returned to the background level (approximatively 3 h).

The graphs reported in Figure 5 show the signals coming from the ethanol, humidity, temperature and CO2 sensors, recorded when the driver was in different conditions, i.e., before drinking alcoholic beverages and therefore in the absence of alcohol in the breath (white zone, left column) and subsequently after drinking an alcoholic beverage and thus in the presence of alcohol in the breath (red zone, right column).

**Figure 5.** Signals from the ethanol, humidity, temperature and CO2 sensors, recorded before drinking alcoholic beverages (white zone, left column) and subsequently after drinking an alcoholic beverage (red zone, right column).

By analyzing data coming from the ethanol sensor, we can see that after the alcoholic beverage is consumed, the signal of the sensor undergoes a quick increase, reaching a maximum after about 30 min (see Figure 6). Subsequently, the signal of the ethanol sensor tends to decrease, as expected by considering the well-known dynamic process of ethanol absorption, metabolism and elimination from the body after its ingestion [17].

The measurements carried out demonstrate that the designed and built sensor module correctly fulfills its functions and is thus able to monitor the level of ethanol in the driver's breath in real time.

#### **4. Conclusions**

An in-vehicle driver breath ethanol detection system was realized by using a simple α-Fe2O3 film-based conductometric sensor for detecting breath ethanol. Using an array provided with the developed ethanol sensor and humidity, temperature and CO2 sensors, it was possible to differentiate the signal of a driver's breath before and after alcohol consumption, thus demonstrating that the developed sensor module can monitor the level of ethanol in the driver's breath in real time.

**Author Contributions:** Conceptualization, G.N. and N.D.; investigation, R.D.C. and M.G.; data curation, R.D.C. and M.G.; writing—review and editing, G.N. and R.D.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by MUR (project code ARS01\_00459) under the framework of *PON*—National Operational Programme for Research and Competitiveness 2014–2020.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


#### *Proceeding Paper*
