3.2.1. Breast Cancer

Electrical impedance spectroscopy/cytometry is a technique that allows the measurement of AC electrical properties of particles in a liquid suspension. This method provides information about the frequency-dependent dielectric parameters of the particles. The main advantage of impedance cytometry is its label-free nature, allowing analysis to be conducted at the individual cell level [41]. To demonstrate, a study by Ahuja et al. [105] presented a microfluidic device that utilizes multifrequency impedance spectroscopy and supervised machine learning, which is shown in Figure 8A, to rapidly evaluate the tumor cell's sensitivity to drugs. In this experiment, T47D cancer cells, which are a type of breast cancer cell, were passed through a microfluidic chip and their impedance and phase features were recorded. The goal of this experiment was to classify T47D cancer cells treated with the target drug and T47D dead cancer cells. The resulting classifier exhibited an accuracy of 95.9% using amplitude change and phase change as features for the SVM classifier.

A surface acoustic wave (SAW) biosensor is an electrical biosensor. It operates based on the generation and detection of surface acoustic waves on a piezoelectric substrate, which are electrical signals. Sountharrajan et al. [106] developed a SAW biosensor for the label-free detection of HER-2/neu, a biomarker associated with breast cancer cells. The biosensor output, along with data from the Wisconsin dataset (the name of the breast cancer dataset), was inputted into a proposed system for data mining classification algorithms. The proposed model was improved by ranking the attributes using the Ranker algorithm, resulting in an accuracy of 79.25% using an SVM classifier. Overall, the study demonstrated the potential of SAW biosensors for the efficient detection of HER-2/neu, offering a promising avenue for early breast cancer diagnosis.

Breast cancer causes metabolic alteration, and volatile metabolites in the breath of patients may be used to diagnose breast cancer [107]. As a proof of concept, Yang et al. [107] developed a new breath test for breast cancer by analyzing the volatile metabolites in exhaled breath (Figure 8B). They collected air samples from breast cancer patients and non-cancer controls and used an electronic nose made of 32 carbon nanotube sensors to analyze the volatile metabolites. Machine learning techniques were employed to create predictive models for breast cancer. Using a RF algorithm, they achieved a 91% accuracy in predicting breast cancer in the test set.

**Figure 8.** (**A**) Multifrequency impedance cytometry measures the response across a broad range of frequencies to assess cellular responses to a target drug. Machine learning algorithms are utilized to predict the viability of both live and dead cells. Reprinted with permission from [105]. Copyright 2021 Springer Nature. (**B**) Graphical representation illustrating the concept of breath biopsy. Breast cancer cells produce volatile metabolites that travel to the lungs and are exhaled. By using a sensor array to analyze these biomarkers in the breath, we can identify the molecular subtype of breast cancer at an early stage. Reprinted from [107]. (**C**) The proposed breast cancer detection system is a Smart Bra. Reprinted with permission from [108]. Copyright 2020 John Wiley and Sons. (**D**) The ML-assisted biochip performs single-cell classification in a label-free manner. The machine learning algorithm is used to perform both cell health classification (cancerous vs. non-cancerous) and cancer subtype cell discrimination at the single cell level. Reprinted with the permission from [109]. Copyright 2020 John Wiley and Sons.

One innovative way to detect breast cancer is through the use of a wearable system designed for detecting breast tumors. For instance, Elsheakh et al. [108] presented a breast cancer detection and monitoring system that utilizes microwave textile-based antenna sensors. The system consists of a wearable device that integrates the microwave antenna sensors and a portable measurement unit that wirelessly communicates with the device to collect and analyze the sensor data, as seen in Figure 8C. The proposed system aims to provide a low-cost, non-invasive, and reliable solution for the early detection and monitoring of breast cancer. The proposed system was tested on a dataset of 110 breast tissue samples, and it achieved an accuracy of 100% for breast cancer detection and classification.

As we discussed earlier, the most commonly used parameter for identifying and quantifying cells is impedance. Joshi et al. [109] demonstrated the effectiveness of single-cell impedance spectroscopy in distinguishing different types of breast cancer cells when used in conjunction with a machine learning classifier (Figure 8D). To evaluate the effectiveness of the method, the researchers pumped two types of cells through a microfluidic channel while constantly measuring the channel's impedance throughout the entire test. The impedance measurements were then fed into a quadratic discriminant analysis (QDA) classifier, which was able to distinguish between the two types of cells with an accuracy of greater than 95% using single-feature classification. As another example, Bondancia et al. [110] developed an immunosensor to detect the cancer biomarker p53 in MCF7 breast cancer cells using electrical impedance spectroscopy. In this sensor, interdigitated electrodes were printed on bacterial nanocellulose substrates using a screen-printing technique. These electrodes were then coated with a layer-by-layer matrix of chitosan and chondroitin sulfate. On top of this matrix, a layer of anti-p53 antibodies was applied by adsorption. They also applied the DT algorithm and achieved 90% accuracy.

To enhance the detection of cancer cells, it is more logical to integrate the electrical and optical methods together. For example, Liang et al. [111] introduced a novel imaging and impedance-based single-cell analysis system called IM2Cell that enables multi-stress level mechanical phenotyping. The system is capable of simultaneously measuring both the mechanical and electrical properties of cells, providing high-dimensional information on cell structures and functions. The authors validated the imaging and impedancebased analyses separately and then combined the techniques to obtain high accuracy in predicting the characteristics of fixed and living MDA-MB-231 breast cancer cells. The authors also demonstrated IM2Cell's ability to classify a mixture of unlabeled MCF-10A, MCF-7, and MDA-MB-231 cell lines with high accuracy. Next, IM2Cell demonstrates a 91.2% classification accuracy in a mixture of unlabeled MCF-10A, MCF-7, and MDA-MB-231 cell lines.

#### 3.2.2. Lung Cancer

In this section, we will explain two examples of different electrical sensors to detect lung cancer cells. The first example is Zhang et al. [112] developing a new biosensing strategy called SHARK (Synthetic Enzyme Shift RNA Signal Amplifier Related Cas13a Knockdown Reaction) for lung cancer detection. SHARK has broad compatibility and can be used as a portable SARS-CoV-2 biosensor with high sensitivity and selectivity, consistent with qRT-PCR results. They combined the output from the biosensors with SVM machine-learning algorithms to predict target miRNAs for (non-small cell lung cancer) NSCLC diagnosis with an accuracy of 82.81%. As another example, Van de Goor et al. [113] utilized five e-nose devices to collect breath samples from lung cancer patients and healthy controls. A total of 60 lung cancer patients and 107 healthy individuals exhaled through the e-nose for five minutes, with the participants assigned to either a training or a blinded control group. The results showed that the e-nose had a diagnostic accuracy of 83%, with a sensitivity of 83%, for discriminating between lung cancer patients and healthy controls. This study provides evidence for the feasibility and effectiveness of using a portable e-nose for accurately detecting lung cancer.

#### 3.2.3. Liver Cancer

Volatile organic compounds (VOCs) in breath are increasingly being recognized as favorable biomarkers, particularly for cancers, due to their ease of sample retrieval and specific association with early metabolic changes [114]. In the article by Nazir and Abbas [114], the use of an e-nose biosensor to detect phenol 2,2-methylene bis, 6 [1,1-D] in breath samples of hepatocellular carcinoma (HCC), which is a type of primary liver cancer, is described. Figure 9 represents an overview of the proposed model.

**Figure 9.** Overview of e-nose biosensor for liver cancer detection from VOCs in breath. Reprinted with permission from [114]. Copyright 2023 Elsevier.

They conducted a screening of breath samples from patients with HCC to identify volatile organic compounds (VOCs) using gas chromatography-mass spectrometry (GC-MS). They applied unsupervised machine learning models to validate their findings. The accuracy of the developed sensor was found to be 86%, demonstrating the promising potential of this approach.

#### 3.2.4. Pancreatic Cancer

Multifrequency single-cell impedance cytometry provides multiparametric biophysical information. To demonstrate, Salahi et al. [115] developed a label-free approach to distinguish pancreatic cancer cells from their associated fibroblasts based on their biophysical properties using impedance cytometry data and machine learning algorithms. The authors demonstrate that gemcitabine treatment changes the biophysical properties of cancer cells and fibroblasts in different ways, resulting in distinguishable patterns in the impedance measurements. The approach has potential applications in cancer diagnosis, treatment monitoring, and drug development.

Combining various types of machine learning techniques has the potential to improve the accuracy of classification. By integrating different approaches, we can leverage the strengths of each method and mitigate their individual limitations, resulting in more precise and reliable classification outcomes. For instance, Honrado et al. [116] improved the classification of cancerous pancreatic cells by combining unsupervised clustering with KNN classification to detect the state of cell death experienced by the cancerous cell. The researchers collected impedance data from flow cytometry and fed it into an unsupervised clustering algorithm that operated at a hyper-dimensional level to autonomously cluster the data. The resulting metrics were then used to quantify the drug-sensitive phenotypes of cancer cells across their progression from viable to early apoptotic, late apoptotic, and necrotic subpopulations. To validate their findings, the team compared the results to those obtained through staining and found that their model was 98.4% accurate in detecting the correct phase of apoptosis in pancreatic cancer cells.
