*Article* **A Maturity Estimation of Bell Pepper (***Capsicum annuum* **L.) by Artificial Vision System for Quality Control**

**Marcos-Jesús Villaseñor-Aguilar 1,2, Micael-Gerardo Bravo-Sánchez 1, José-Alfredo Padilla-Medina 1, Jorge Luis Vázquez-Vera 1, Ramón-Gerardo Guevara-González 3, Francisco-Javier García-Rodríguez <sup>1</sup> and Alejandro-Israel Barranco-Gutiérrez 1,4,\***


Received: 4 June 2020; Accepted: 21 July 2020; Published: 24 July 2020

**Abstract:** Sweet bell peppers are a Solanaceous fruit belonging to the*Capsicum annuum*L. species whose consumption is popular in world gastronomy due to its wide variety of colors (ranging green, yellow, orange, red, and purple), shapes, and sizes and the absence of spicy flavor. In addition, these fruits have a characteristic flavor and nutritional attributes that include ascorbic acid, polyphenols, and carotenoids. A quality criterion for the harvest of this fruit is maturity; this attribute is visually determined by the consumer when verifying the color of the fruit's pericarp. The present work proposes an artificial vision system that automatically describes ripeness levels of the bell pepper and compares the Fuzzy logic (FL) and Neuronal Networks for the classification stage. In this investigation, maturity stages of bell peppers were referenced by measuring total soluble solids (TSS), ◦ Brix, using refractometry. The proposed method was integrated in four stages. The first one consists in the image acquisition of five views using the Raspberry Pi 5 Megapixel camera. The second one is the segmentation of acquired image samples, where background and noise are removed from each image. The third phase is the segmentation of the regions of interest (green, yellow, orange and red) using the connect components algorithm to select areas. The last phase is the classification, which outputs the maturity stage. The classificatory was designed using Matlab's Fuzzy Logic Toolbox and Deep Learning Toolbox. Its implementation was carried out onto Raspberry Pi platform. It tested the maturity classifier models using neural networks (RBF-ANN) and fuzzy logic models (ANFIS) with an accuracy of 100% and 88%, respectively. Finally, it was constructed with a content of ◦ Brix prediction model with small improvements regarding the state of art.

**Keywords:** bell pepper; maturity; fuzzy logic; computational vision

### **1. Introduction**

Native to the Americas, sweet bell peppers are a Solanaceous fruit belonging to the *Capsicum annuum* L. species. It is a non-pungent fruit that is valued for its color, flavor, and nutritional attributes including ascorbic acid, polyphenolics, and various carotenoids. It comes in a wide variety of colors (ranging green, yellow, orange, red, and purple), shapes, and sizes, as well as because it has a

high content of ascorbic acid, polyphenols, and other antioxidants. Nowadays, bell peppers are widely consumed in various ways, dehydrated, preserved, frozen, or raw for packaged salads. Generally, the harvest of bell peppers is determined by the size, color, and texture of the fruit. Traditionally, the harvest of this fruit is done by reaching physiological mute when the pericarp becomes thick and the fruit reaches the typical size. However, estimating pepper maturity at the green stage can be difficult even for fruit with similar physical attributes [1]. Under certain conditions, bell peppers can begin to ripen during shipping. Partially ripened fruit, classified as chocolate or suntan, have lower market values than at the solid color stage.

The bell peppers reach their optimum state of maturity for use in the kitchen when they are in solid color. Consumers prefer this fruit in its best stage of maturity more so than its physical appearance and nutritional content [2]. Dutch researchers specializing in the sensory area, reported that study groups have considered that more ripeness of bell peppers is sweeter and has a red pepper aroma, while those in the green stage were rated for bitterness and aroma of herbs and cucumber [3]. On the other hand, ripe peppers are more expensive to produce, due to the longer time required for ripening and the greater likelihood of damage from insects or disease. Furthermore, ripe peppers are more susceptible to physical damage during transport. In addition, they have a shorter shelf life due to the stage of maturity at which they are harvested. These fruits are non-climacteric regarding postharvest respiratory patterns. At the mentioned stage, bell peppers will progress through the normal ripening process to degrade chlorophyll while simultaneously synthesizing a variety of red and yellow carotenoids. The red and yellow varieties are the peppers most on demand by consumers, followed by the orange and purple varieties [4].

Bell pepper contains provitamin A, carotenoids, and xanthophylls. Many studies have focused on improving retention of these compounds during processing and storage [5–10] that increase their concentration as the fruit reaches a major state of maturity. Bell peppers also contain high concentrations of ascorbic acid (0.15–2.0 mg·g−<sup>1</sup> fresh weight) compared to other fruits and vegetables [11–30]. The production of ascorbic acid in peppers and other fruits are related to glucose metabolism and light exposure, and concentrations of both ascorbic acid and sugar reduction typically increases with the stage of maturity [14]. Polyphenolics are also an important chemical component in bell peppers and impart functional properties to the plant such as disease resistance and potential health benefits to consumers. Some studies found that total phenolics, including the flavonoid quercetin, decrease with increased maturity for yellow bell peppers, but increased for other pepper varieties [6]. Estrada et al. [31] also demonstrated a decrease in free phenolic concentrations in peppers over five stages of maturity. Pepper growers could reduce field production costs by hastening the fruit ripening rate on the plant or by harvesting the fruit before attaining full color and completing the ripening process during storage without appreciable loss in quality or phytochemical attributes. An important criterion for consumers of sweet pepper is its sweetness, commonly estimated with the content of soluble solids (SSC) [22]. This parameter is associated with the different stages of fruit maturity [24–28]. The measurement of SSCs is traditionally a destructive test, to perform it, the most used instrument is the refractometer. This is an optical instrument that measures the refractive index of the juice from the sample [29,30].

At present, the food industry has demanded the use of non-invasive high precision measuring equipment that determines the external and internal quality parameters of the fruits [32]. In this respect, one of the disciplines that has had a great impact on fruit quality control is computational vision (VC). This has the characteristic that emulates the functionality of human vision and allows spatial and optical information from the captured image of the sample. Different investigations have been reported with VC focused on determining the degrees of ripeness of various fruits such as persimmon, strawberries, pomegranate, and tomato [33–35]. The present work aims to implement an artificial vision system that automatically describes the ripeness levels of the bell pepper.

More recently, total soluble solids (TSS) or Brix grades are a classical tool to determine the maturity of the fruits in the food industry even though it is a destructive technique. The content of TSS consists of 80–95% sugars and the measure of TSS is associated with the dissolved sugars in cell juice [36]. These authors affirmed that the quantity of sugars in the fruit depends mainly on the variety, the assimilatory yield of the leaves, the leaf/fruit ratio, the climatic conditions during the development of the fruit, the state of development, and the maturity. The accumulation of sugars is associated with the development of optimum quality for consumption. Although the sugars can be transported to the fruit by the sap, they are also contributed by the splitting of the starch reserves of the fruits [36]. When the fruits in general have their highest sugar content, they have reached their physiological maturity, which coincides with what was investigated in relation to the bell peppers and their greater ripeness [37]. The TSS content showed a constant increase as the fruit maturity status increased, which could be seen with an increase in color fastness in the sample bell peppers. The ascending behavior of TSS content is consistent with that reported in the literature [37]. Kays [38] explained that, when the fruit is ripening in the plant, the sugars increase their concentration by the translocation of sucrose from the leaves, which occurs in most species. However, there is also the recycling of the respiratory substrate from the carbon stored in the fruit. Overall, highly significant positive correlations of soluble solids content were found with this research. The measurement of the TSS was done in triplicate with two different refractometers, an ATAGO Refractometer (MASTER-M model, Tokyo, Japan), Brix measurement scale of 0.0–33.0%, with a minimum of 0.2%, an accuracy of ±0.2%, and repeatability of ±0.1%; and a HANNA brand refractometer, model HI 96802 (Woonsocket, RI, USA), Brix measurement scale of 0.0–85.0%, with a temperature range of 0–80 ◦C, an accuracy of ±0.2%, and repeatability of ±0.1%.

Figure 1 shows the distribution of Brix grades of soluble solids content in the four stages of maturity of the fruit. This allowed us to establish that the grouping of the samples contemplates different stages of maturity of the bell pepper. This describes an increase in sugars from maturity Stage 1 to Stage 3 and the maximum sugar content is in State 4.

**Figure 1.** Distribution of Brix Grades in different stages of maturity.

### **2. Related Work**

Currently, the high demand in the food industry has created the need for a fast and effective way to control the quality of products, which has led to the application of intelligent systems such as artificial vision that helps determine the internal and external properties of the fruits resulting in the stage of maturity. Different authors have focused on determining the maturity of various fruits such as apricot, nightshade, citrus, coffee, and mango. It is worth mentioning several studies that have focused on determining the maturity of bell peppers and estimating total soluble solids [39].

Harel et al. [40] classified their maturity using the random forest algorithm, which identified four classes: fully green, partially colored immature, mature partially colored, and colored.

Elhariri et al. [41] used an algorithm based on a support vector machine (SVM) for classification. This identified five maturity classes associated with green and red shades.

Shamili et al. [42] proposed a polynomial regression model that estimated the soluble solids content of mango. The descriptor that he proposed was the normal values of \* a of the images captured from five types of mangoes; the samples he used had different percentages of skin coloration of 0%, 20%, 25%, 50%, and more than 50%.

Li et al. [43] used hyperspectral imaging in the visible and near-infrared (VNIR) and short-wave near-infrared (SWIR) regions focused on measuring maturity, firmness, and suspended sediment concentration (SSC). This research showed that there is a strong correlation between the SSC and the average spectra obtained from one or two opposite sides of the fruit in the SWIR region.

Another similar work was carried out by Teerachaichayut et al. [44], who proposed several models for limes to determine total soluble solids (TSS) and titratable acidity (TA) using partial least square regression.

### **3. Materials and Methods**

### *3.1. Samples*

A sample set of 50 bell peppers produced in the Laja-Bajió region, Guanajuato, Mexico was analyzed. The selected sample attributes consider different maturity degrees and homogeneous size [45]. For the training of the vision system, fruits with homogeneous and defined colors were chosen. Therefore, samples showing heterogeneous colors with various spots in the pericarp were excluded for this stage but were included for general testing. The bell peppers were classified into four classes, as shown in Figure 2. Class 1 grouped ten green samples. Class 2 was made up of seven yellow fruits. Class 3 was fourteen orange pigmented fruits. In Class 4, nineteen samples with a red hue were grouped.

**Figure 2.** Samples of bell peppers in different degrees of maturity.

### *3.2. General Structure of Artificial Vision System*

Figure 3 shows the proposed method for predicting the sugar content of bell pepper. The first step of the proposed method was the acquisition of the image samples using computer vision system (VCS). The second step was the segmentation of the fruit in the image, where the background and the supersaturated pixels were removed. The third step was the masking step to identify the regions of interest with green (GPOI), yellow (YPOI), orange (OPOI), and red (RPOI) pixels. These were obtained by means of four masks that use the red, green, blue (RGB) components of the segmented images. The fourth step was to obtain the areas of each region of interest (GAROI, YAROI, OAROI and RAROI). The fifth step was the classification of the fruit with the areas from the filters. The last step was the prediction of ◦ Brix, which was done with the class and the areas determined by the masks.

**Figure 3.** Proposed method to classify degrees of maturity in the bell pepper.

### *3.3. Image Acquisition*

Five images per fruit were captured, as shown in Figure 4. The image format was JPG with a resolution of 768 × 1366 × 3. The acquisition was made with a computational vision system (VSC) that operates with OpenCV-Python language. This was integrated by the isolation, lighting, image capture, and image processing subsystems. The insulation subsystem consisted of a black cabinet of dimensions 38 cm × 38 cm × 43 cm, which allowed the reduction of variations in lighting. The architecture used by the lighting subsystem was a 5.4 W led ring configuration placed 30 cm above sample. The image capture was performed with Raspberry camera module (8 megapixels), and the processing platform was Raspberry Pi 3 card [33].

**Figure 4.** Images acquired by vision system.

### *3.4. Automatic Sample Classification*

In the second phase, the degrees and averages of the RGB channels of the segmented images of each fruit were mapped, as shown in Figure 5. G (green) corresponds to samples belonging to Class 1, Y (yellow) label to samples of Class 2, O (orange) label to samples of Class 3, and R (red) label to the samples of Class 4. In the last phase, the regions of interest corresponding to the green, yellow, orange, and red tones were carried out.

**Figure 5.** Mapping averages of the red, green, blue (RGB) channels for four classes of bell peppers.

### *3.5. Obtaining Regions of Interest*

Obtaining regions of interest used the methodology proposed by Goel et al [32], as shown in Figure 6. (i) The first step required the binarization of the captured images, which was done using the Hue, Saturation, Value (HSV) color space model. This change in the color space model allowed obtaining the frequency of each color in the visible spectrum with component H, the purity of color with component S, and the proximity of the pixel to black and white with component V. The thresholds used to segment the background of the samples were 122 ≤ H ≤ 255, 127 ≤ S ≤ 255, and −0 ≤ V ≤ 255 using a range from 0 to 255. (ii) The second step was segmentation of images of each view of bell pepper using a component connection algorithm, and later all the small regions smaller than 400 pixels that did not correspond to the binarized image of the sample were discriminated. (iii) The last step was to use respective masks to obtain regions of interest due to its color. Table 1 shows the four ranges of the Hue parameters to identify the pixel areas with green (47 ≤ H ≤ 118), yellow (32 ≤ H ≤ 46), orange (15 ≤ H ≤ 32), and red (155 ≤ H ≤ 241) hue of the images. Most of the masks except the red mask one used the range from 0 to 255 for the S and V components to identify the different color regions. The red mask used a smaller range of the S component that allowed the areas with orange and red pixels to be correctly identified. These thresholds were obtained using the 50 samples where each of its segmented images was analyzed. These masks identify the pixels corresponding to the green, yellow, orange, and red shades.

**Figure 6.** Pre-processing of images: (**i**) binarization of the image captured from the fruit; (**ii**) segmentation of the image and discrimination of areas; (**iii**) obtaining segmented sample applying the green mask of interest region; (**iv**) obtaining segmented sample applying the yellow mask of interest region; (**v**) obtaining segmented sample applying the orange mask of interest region; and (**vi**) obtaining segmented sample applying the red mask of interest region.



### *3.6. Maturity Status Estimator*

### 3.6.1. Artificial Neural Network (ANN)

Artificial neural networks (ANN) are mathematical models that emulate the functioning of biological neural networks. This was used in this study for its structural simplicity, learning skills, and its application for the approach, classification, and pattern recognition. Figure 7 presents the ANN architecture that was used to estimate the ◦ Brix content of the fifty samples presented in Figure 1. Its architecture consisted of an input layer, three hidden layers, and an output layer. The input layer is used to present the training and test patterns that correspond to the GAROI, YAROI, OAROI, and RAROI regions of interest. The hidden first layer used radial base-type activation functions, each of its neurons calculating the similarity between the input and its training set. The second hidden layer adds the values of each neuron from the first hidden layer that are multiplied by their weight associated with each neuron to obtain the class that the sample belongs to. The third layer of neurons employs neurons with sigmoidal-type firing functions where the sample classification information and the four regions of interest were weighted. The output layer is of the linear type, allowing the Brix content to be estimated using the weights associated with the neurons of the third hidden layer.

**Figure 7.** Brix prediction Artificial neural networks (ANN) model.

One of the ANNs most used in classification tasks is radial basis function ANN (RBF-ANN), as shown in Figure 8. Its architecture uses an input layer, hidden layers, and output layer. The input layer is used to present the training and test patterns. The hidden layer is made up of radial (Gaussian) functions that are completely interconnected between all its nodes with the input layer, which are activated by this function. The output layer is activated by the linear functions that determine the classification and is interconnected with all the nodes of the hidden layer.

**Figure 8.** RBF-ANN maturity classifier.

The proposed RBFNN classification models were developed with Matlab's Deep Learning Toolbox. The main difference between them is the number of neurons that the hidden layer contains, as shown in Table 2. The inputs used by each classifier are the interest regions with shades of green, yellow, orange, and red and the outputs are the four classes. The model design used 70% of the data for training, 20% for validation, and the rest for the testing stage.


**Table 2.** RBF-ANN maturity classifier.

Together, three models of two-layer feed-forward network with sigmoid hidden neurons and linear output neurons (Fitnet) are proposed. Figure 9 presents the architecture of the implemented model to predict the ◦ Brix content of bell peppers. Three models were proposed where their main difference is the number of neurons in the hidden layer, as shown in Table 3; the precision of each model to predict the ◦ Brix was different.

**Figure 9.** Architectural description of the model for ◦ Brix prediction using Fitnet.


**Table 3.** Models using neural network Feedback prediction of Brix degrees.

### 3.6.2. Fuzzy Logic

Fuzzy Logic (FL) is a discipline of Artificial Intelligence that analyzes real-world information on a scale between true and false. This is integrated by three stages: fuzzification, the inference that uses a series of linguistic rules, and defuzzification. Figure 10 shows a structure of the fuzzy classification system proposed to classify bell peppers. It has four inputs and one output.

The regions of interest of each shade correspond to the input and are used to identify the degree of maturity associated with the output fuzzy system. It is used to identify segments or regions of interest with shades of green, yellow, orange, and red. The output has different ranges for each class.

Fuzzification of the input variables was carried out with membership functions of triangular type, which are characterized by their easy implementation in hardware. The linguistic variables used were low, medium, and high, as shown in Figures 11–14. The mathematical functions of each membership function are described in Equations (1)–(12). Together, the different pixel areas were used to increase the sensitivity of the fuzzy system to infer maturity changes.

$$\text{LowGARO} = \left| \frac{50 - \text{GAROI}}{50} 0 \right.\\ \left. \text{$$

$$\text{Galaxum} \text{GAROI} [\frac{\text{GAROI}}{50} 0 < \text{GAROI} \le 50 \,\, \frac{100 - \text{GAROI}}{50} 50 < \text{GAROI} \le 100 \,\tag{2}$$

$$\text{GHighGARM} = |00 \times \text{GARM} \le 50 \, \frac{\text{GARM} - 50}{50} \text{50} < \text{GARM} \le 100 \tag{3}$$

$$\text{LowYAROI} = \left| \frac{29.09 - \text{YAROI}}{29.09} 0 < \text{YAROI} \le 16.48 \, 016.48 < \text{YAROI} \le 34.73 \right. \tag{4}$$

$$\text{MaliumYAROI} = |\frac{\text{YAROI}}{29.09} 0 < \text{YAROI} \le 29.09 \,\frac{58.17 - \text{YAROI}}{29.08} V < \text{YAROI} \le 58.17\tag{5}$$

$$YARIARII = 00 < YARIOI \le 29.09 \,\, \frac{YARIOI - 29.09}{29.09} 29.09 < YARIOI \le 58.17\tag{6}$$

$$\text{LowOAROII} = (\frac{16.48 - OAROII}{16.48} 0 < OAROII \le 16.48 \, \frac{OAROII - 34.73}{18.25} 16.48 < OAROII \le 34.37 \tag{7}$$

$$\text{MediumOAROII} = \left\{ \frac{16.48 - \text{OAROII}}{16.48} 0 < \text{OAROII} \le 16.48 \right\} \frac{\text{OAROII} - 34.73}{18.25} 16.48 < \text{OAROII} \le 34.37 \quad \text{(8)}$$

$$\text{MighOAROII} = \left| \frac{16.48 - \text{OAROII}}{16.48} 0 < \text{OAROII} \le 16.48 \right.\\ \frac{\text{OAROII} - 34.73}{18.25} 16.48 < \text{OAROII} \le 34.37 \tag{9}$$

$$Llow\ RROI = \begin{cases} \frac{16.48 - RROI}{16.48} & 0 < RROI \le 16.48\\ \frac{RROI - 34.73}{18.25} & 16.48 < RROI \le 34.37 \end{cases} \tag{10}$$

$$Medium\\_RROI = \begin{cases} \frac{16.48 - RROI}{16.48} & 0 < RROI \le 16.48\\ \frac{RROI - 34.73}{18.25} & 16.48 < RROI \le 34.37 \end{cases} \tag{11}$$

$$R \text{ High } RROI = \begin{cases} \frac{16.48 - RROI}{16.48} & 0 < RROI \le 16.48\\ \frac{RROI - 34.73}{18.25} & 16.48 < RROI \le 34.37 \end{cases} \tag{12}$$

**Figure 12.** Membership functions of YAROI input.

**Figure 14.** Membership functions of RAROI input.

Figure 15 shows the fuzzy Takagi–Sugeno model used for the development of the classifier; this was selected for its low computational cost unlike the Mandami model. It uses 81 inference rules

obtained from using fuzzy values of interest regions. The transforming fuzzy values of classification output was carried out using Equation (13). The final output is determined by rules using *Zi* levels of output and weight *wi* of the rule.

$$\text{Final Output} = \frac{\sum\_{i=1}^{N} w\_i Z\_i}{\sum\_{i=1}^{18} w\_i} \tag{13}$$

**Figure 15.** Operation of Takagi–Sugeno rules to classify maturity of bell peppers.

Five fuzzy classification models were designed using the Matlab R2014a Fuzzy Logic Toolbox. Its training used a set of 50 vectors consisting of areas the interest regions in shades of green, yellow, orange, and red and their label corresponding to their maturity. The training used 10 epochs. The main difference between the models is the number of membership functions. Table 4 shows the number of membership functions used by each model and their training error.


**Table 4.** Classification models.

Three models were proposed where their main difference is the number of neurons in the hidden layer, as shown in Table 5; the precision of each model to predict the Brix degrees was different.


**Table 5.** Models using FL prediction of ◦ Brix.

### **4. Results**

Figure 16 shows the accuracy of the 10 maturity classifier models using ANN and FL. The first five models are type ANN (blue circles) and the next five models are type FL (red squares). It can be seen that the proposed models which used ANN achieved greater than 90% accuracy, unlike those that used FL. Together, the models that had good precision are Models 13–15. Finally, the worst models were Models 11 and 12 with an accuracy of less than 70%. In Figure 17, the prediction error of the model is presented, where it can be highlighted that the models which use ANN have lower least squared medium error. Five ANN models (blue circles) and three FL models (red squares) are shown.

**Figure 16.** Comparison of maturity prediction models.

**Figure 17.** Comparison of ◦ Brix prediction models.

### **5. Discussion**

The main contribution of this work is the development of a ◦ Brix content prediction system for bell pepper maturity. According to the results, the models with FL achieved a maximum precision of 88% in identifying the four stages of maturity corresponding to the shades of green, yellow, orange, and red. The models with ANN have 100% precision to identify samples of green and red color similar to the results reported by Elhariri et al. [41]. Of the results obtained, Model 8 was the one that presented a correlation of R = 0.79543 between the green, yellow, orange, and red regions of interest and ◦ Brix. This result is similar to that reported by Shamili [42] and is slightly superior to those reported by Leiva-Valenzuela et al. [46] (R = 0.788) and Rahman et al. [47] (R = 0.74). The advantage is that our proposal uses a visible RGB camera and they used a high-cost multispectral camera. Figure 18 shows the estimation.

**Figure 18.** The samples ◦ Brix was measured with refractometry. The measurements are presented as circles and the proposed system's estimates are indicated by squares.

### **6. Conclusions**

In this work, a new classification system is proposed to evaluate the bell peppers maturity using an artificial vision system. According to the results obtained, it can be concluded that the proposed architecture improves the ◦ Brix prediction using the regions of interest associated with the color of the pericarp. The correlation obtained was R = 0.7929 with the use of the NETFIT-ANN, surpassing the model designed with FL with a correlation of R = 0.696. Together, it was possible to classify 100% of the classes of the samples with the use of a model that uses the RBF-ANN architecture. This architecture presented a better result than the FL models that obtained a maximum accuracy of 88%. The present work demonstrated that it is possible to identify the degrees of maturity of the peppers using an artificial vision system that is sensitive to the total soluble solids content of the fruit. It has the advantage of using a low-cost RGB camera instead of a multispectral one, and it is a non-destructive technique to estimate ◦ Brix. This vision system model is applicable to real scenarios in the industrial sector such as online processes. Furthermore, it is a system that can be used to predict the ◦ Brix content in real time. As further research, the development of a prediction system for ascorbic acid, polyphenolics, and various carotenoids has been completed.

**Author Contributions:** Conceptualization, M.-J.V.-A., M.-G.B.-S. and A.-I.B.-G.; methodology, J.-A.P.-M.; software, M.-J.V.-A.; validation, M.-G.B.-S.; formal analysis, J.L.V.-V.; investigation, R.-G.G.-G.; data curation, F.-J.G.-R.; writing—original draft preparation, M.-G.B.-S. and M.-J.V.-A.; writing—review and editing, A.-I.B.-G.; and supervision, A.-I.B.-G. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by CONACyT and Tecnológico Nacional de Mexico.

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

**Link to Images Database:** https://drive.google.com/drive/u/0/folders/1gzmhn1E16IPGkMnHX8yOfp2FjUbgNsW6

### **References**


© 2020 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 (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Emerging Trends in Research on Food Compounds and Women's Fertility: A Systematic Review**

### **Aleksandra Bykowska-Derda 1, Ezgi Kolay 1, Malgorzata Kaluzna <sup>2</sup> and Magdalena Czlapka-Matyasik 1,\***


Received: 28 May 2020; Accepted: 24 June 2020; Published: 29 June 2020

### **Featured Application: The application of this systematic review is a comprehensive study of food compounds, nutrition, food production, health and environmental sciences in improving female fertility.**

**Abstract:** Pro-healthy behaviours, including the diet, are significant factors in maintaining women's fertility health. However, to improve the patient's nutrition management, it is important to seek food-derived bioactive compounds to support fertility treatment. This review analysed recent studies of food compounds related to fertility, using databases including PubMed, Web of Science and Science Direct as well as PRISMA (preferred reporting items for systematic reviews) to ensure complete and transparent reporting of systematic reviews. This review lists foods associated with a higher birth rate, using original papers from the last five years (2015). The analysis included the impact of food compounds such as caffeine, fatty acids, folates and vitamin D, as well as the intake of fish, whole grains, dairy and soya. In addition, dietary patterns and total diet composition supporting women's fertility were also analysed. The results will encourage further research on the relationship between food components and fertility.

**Keywords:** female infertility; nutrient; vitamin D; folates; soy; antioxidants; minerals; vitamins; food research trends; environmental impact

### **1. Introduction**

Fertility, known as the ability to establish a clinical pregnancy, is dependent on multiple factors, including female age, environmental pollution, diet, tobacco use, alcohol intake, as well as diseases affecting endocrine function and the anatomy of the reproduction system [1,2]. In turn, infertility is medically defined as a failure to establish a clinical pregnancy after 12 months of regular and unprotected sexual intercourse. It is estimated that infertility in women of child-bearing age is 1 in 7 couples in developed countries and 1 in 4 couples in developing countries, which is increasing significantly [1]. The demand for infertility services is still growing and can be improved thanks to technological advances and the development of medicine.

Factors influencing fertility may be unmodifiable, such as age and environment, or could be medically treated, such as health status—including endocrine disorders. Furthermore, some factors could be modifiable, such as health behaviours, dietary patterns and micronutrient intake.

It has been confirmed that in-vitro fertilisation success rate seems to be highest during the summer months when the pollution of particulate matter (PM) is at its lowest [3]. Phthalates, which may negatively influence the fertility health of women [4], were found to affect the occupational health

of hairdressers [5]. Some primary factors, such as excess body weight or underweight, also decrease the rate of fertility [6,7]. Body saturation with vitamin D was suggested to have a beneficial influence on fertility [8–12].

The recommendations by the Committee of the American Society for Reproductive Medicine in collaboration with the Society for Reproductive Endocrinology and Infertility American Society for Reproductive Medicine, advise females to follow a healthy diet, avoid alcohol and decrease caffeine intake to a moderate level [13]. It is also advisable for women to supplement folate (400 μg/day) to decrease the chance of neural tube defects [13]. However, there is some evidence showing that some food ingredients and specific dietary patterns in women may be positively associated with pregnancy and live birth rates [14–16].

One of the food ingredients widely discussed in the context of fertility is sugar. It was shown that its presence in the diet reduces nutritional density and worsens its nutritional quality [17]. The possible mechanism between sugar-sweetened beverages and fertility was explained by increased insulin resistance, leading to oxidative stress. This relation may deleteriously affect semen quality and ovulatory function. Such a mechanism was hypothesised by Hatz, et al. [18], who studied a group of nearly five thousand women and found that fertility and the amount of sugar consumed in sugar-sweetened beverages (particularly sodas and energy drinks) was associated with lower fecundability [18].

A large group of compounds that are still being studied are bioactive compounds in food. Their roles in oxidative stress and fertility have been presented in many studies [19–22], including several studies on female animal models and bioactive food compounds [23–25]. Their role is hypothesised to diminish the effect on the endocrine system of disruptive chemicals. For example, there is some evidence that animals treated with BPA (Bisphenol A), after maternal supplementation of folate and a high phytoestrogen diet, influence oocyte growth and foetal methylation of DNA [26,27]. Other studies have highlighted that a low dose (but not a high dose) of ginger powder, improved the follicle counts of rats [23]. Therefore, it is not clear what dose will be as effective on humans. Additionally, human fertility is affected by complex and multiple factors which could be difficult to expose animals to.

The time of preconception may motivate couples to adopt healthier behaviours and to seek information on factors improving fertility. Even though medical consultation is still the most common source for seeking advice for fertility, social media and the internet also play significant roles [28]. The choice of supplement options is vast, as the fertility supplement market is continuously growing and it has been estimated to be worth USD 1.45 billion globally in 2018 [29]. The use of supplements is not always recommended by healthcare professionals, and their misuse may even pose a threat to health.

The literature related to food compounds and fertility has not been extensively collected, and there is no consensus on what the trends are in these studies or what groups of food ingredients should be considered as supportive or detrimental to fertility. In light of this evidence, an analysis of diet ingredients, food research (e.g., ginger, BPA) and fertility could lead to new supporting therapy strategies that affect the birth rate through the modulation of eating habits. Accordingly, this review provides an analysis of new food compound research influencing fertility and revises recent studies involving the impact of food bioactive compounds on women's fertility.

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

### *2.1. Search Strategy*

A systematic search of literature published before December 2019 was performed in PubMed (National Institiute of Health, USA)(https://www.ncbi.nlm.nih.gov/pubmed), Web of Science (Clarivate Analytics, USA) (https://www.webofknowledge.com), Scopus (Elsevier, RELX Group plc), (https://www.scopus.com) and Science Direct (Elsevier, RELX Group plc) (https://www.sciencedirect. com/) to identify studies describing the association between bioactive food compound intake and

women's fertility. The search strategy was restricted to English language original articles. The following types of documents were excluded: review, book and book chapters.

The search was based upon the following index terms, titles or abstracts listed below: ((bioactive OR nutrient OR food OR ingredient OR vitamin OR mineral OR antioxidant OR phytonutrient) AND (fertility)). The protocol was registered in the "PROSPERO International prospective register of systematic reviews" PROSPERO 2020: CRD42020160223 and is available on https://www.crd.york.ac. uk/prospero/display\_record.php?ID = CRD42020160223.

### *2.2. Inclusion and Exclusion Criteria*

Studies on the influence of food compounds on infertility, signs and symptom changes in patients affected by infertility were included. Studies using different food components concerning changes in the concentration of biomarkers for the assessment of infertility and changes in symptoms were analysed. The systematic search included a population of women in the reproductive age 21–50 with diagnosed infertility or healthy women trying to conceive. All studies conducted on animals and case reports were excluded. The studies included were both qualitative and quantitative. A quality assessment of questionable articles was performed with a checklist described by Kmet et al. [30]. Articles written in a language other than English were excluded. Since the search included new trends in food research, it only included articles within the last five years (2015).

### *2.3. Study Extraction Process*

The study selection process includes an assessment of articles based upon titles, abstracts and full text, which were performed by two independent researchers in parallel in each database. At each step of the assessment, all disagreements between the researchers were resolved after consultation with the review coordinator. Only in the case of disagreement during the title assessment process was the paper included in the next step. Full-texts of all records that were selected in the abstract review phase were searched for through the library of Poznan University of Life Sciences.

### **3. Results**

A total of 4609 studies were screened for inclusion in this systematic review. After the elimination process (Figure 1), a total of 25 qualitative studies and 4 quantitative studies were included. The studies were performed internationally and included the following countries USA (n = 20) [15,31–45], Australia (n = 1) [46], New Zealand (n = 1) [46], Ireland (n = 1) [46], United Kingdom (n = 2) [46,47], China (n = 1) [48], Denmark (n = 4) [41,43,49,50], Greece (n = 1) [14], Iran (n = 2) [51,52], Brazil (n = 2) [53,54], Canada (n = 1) [43], Russia (n = 1) [55], Italy (n = 2) [56,57], Spain (n = 1) [58]. The articles concerned female fertility and the intake of a Mediterranean dietary pattern (n = 2) [51,57], fruit, vegetables and whole grain intake (n = 3) [38,46,53], fish (n = 1) [42], dairy (n = 3) [15,43, 44], types of fatty acids (n = 4) [39–41], soy (n = 2) [32,33], caffeine (n = 3) [45,50,54], folate and B12 (n = 4) [34,35,59,60], melatonin (n = 1) [58], CoQ10 (n = 1) [48] as well as a combination of different compounds (n = 4) [37,47,52,55]. The women who participated in the studies were either planning pregnancy (n = 6) [36,40,41,45,46,55], infertile (n = 3) [49,51,55], undergoing or subjected to assistive reproductive technology (ART) therapy (n = 16) [14,15,31,34,35,39,42,44,50,52–54,57,58]. The main results of the studies have been summarized in Table 1 (qualitative studies) and Table 2 (quantitative studies).

**Figure 1.** Preferred reporting items for systematic reviews (PRISMA) Study selection process diagram.




**Table 1.** *Cont.*


25(OH)D-*25*-*hydroxyvitamin D, ART—Assistive Reproductive Technologies,* AMH—Anti-Mullerian Hormone, FSH—Follicle-stimulating Hormone, FFQ—Food Frequency Questionnaire, EAR—Estimated Average Requirement, BPA—Bisphenol A, FA—fatty acid, IVF—In Vitro Fertilization, ICSI—Intracytoplasmic Sperm Injection.

**Table 2.** Quantitative studies concerning bioactive food components and female fertility qualified for the review.


IVF—In Vitro Fertilization, ICSI—Intracytoplasmic Sperm Injection.

### **4. Discussion**

This systematic review aimed to identify emergent trends in food compounds studies which influence women's fertility. We qualified a total of 29 studies from the past five years among women planning to conceive either naturally or with assisted reproductive technologies.

### *4.1. Dietary Patterns, Intake of Fruits, Vegetables and Whole Grains*

The modern diet and nutrition analysis based on dietary patterns also assessed the nutrition behaviours as a whole, rather than looking at a single nutrient. Dietary patterns are defined as the quantity, variety, or combination of different foods and beverages in a diet and the frequency in which they are habitually consumed. The frequency reflects food compounds consumed in the diet directly. The most common dietary patterns are pro-healthy, Mediterranean, western and dairy-related [61,62]. The results of this systematic review (presented in Table 1) showed that a high intake of fruits and vegetables and adherence to a pro-healthy dietary pattern is associated with a higher average number of oocytes [51] and embryo quality [53]. The Mediterranean dietary pattern has been associated with supporting fertility health in women. Karayannis et al. found that this dietary pattern is only related to the live birth rate in women under the age of 35 [14]. Another study showed that it was related only to higher oocyte number and clinical pregnancy in women over 35 [57]. The Mediterranean diet is characterised by a high intake of extra virgin olive oil, vegetables, fruits, cereals, nuts and legumes, a moderate intake of fish and other meat, dairy products and red wine and low intakes of eggs and sweets [63]. Another dietary pattern used in the studies was a fertility diet dietary pattern characterised by the intake of supplemental folate and B12, low-pesticide residue produce, high intake of whole grains, seafood, dairy and soy foods. This dietary pattern was positively related to the likelihood of live birth [31].

### *4.2. Fatty Acids and Fish Intake*

The current review has shown that there is no conclusive evidence about the impact of polyunsaturated fatty acids, including the intake of omega-3 on human fertility [13]. The results of the review are inconclusive: two studies found an association between omega-3 fatty acids and fertility [40,41], while another study found no impact of these fatty acids on fertility [39]. However, all of the studies analysing fatty acid intake found that the amount of trans fatty acid consumed is negatively related to the live birth ratio [39,41]. These results also support Grieger et al., who found that a high consumption of fast foods (a rich source of trans fatty acids) influences fertility [46].

### *4.3. Dairy*

For many years, the topic of high dairy intake has been controversial and linked with both a positive and negative impact on the health of women [16,64,65]. Nevertheless, to date, most of the studies concerning female reproductive health have been skewed towards a positive association. Moreover, all of the studies in this systematic search, including the period 2015–2019 concerning dairy intake and fertility, have found at least a small positive relationship between these variables [15,43,44]. It should be noted that dairy foods are generally perceived as a pro-healthy dietary attribute. We studied this group of foods previously and found that a more significant effect on dairy consumption by women was the family environment than health-related protective factors [66].

### *4.4. Ca*ff*eine*

Decreasing the caffeine intake to moderate during the time of preconception has been recommended to couples who plan pregnancy [13]. A high intake of caffeine, especially black tea [45] and coffee with added sugar and diet soft drinks [54] has been related to a lower live birth rate. However, coffee intake of 1–5 cups daily has been associated with higher chances of a live birth than none [50]. The intake of caffeine and coffee may be associated with more favourable dietary patterns and health behaviours which could influence fertility [67]. This could be the reason why different types of caffeinated beverages bring contrasting results.

### *4.5. Soy*

The use of nutrients as factor diminishing adverse health effects of environmental pollutants was found in research concerning soy. Soy, as the food product containing phytosterols, could alleviate the effect of endocrine disruptor bisphenol A. This hypothesis is supported by a study concerning women undergoing ART, urinary bisphenol A and soy intake and their influence on live birth rate [32]. Intake of soya, regardless of environmental pollutants, was also related to a higher probability of life birth [33]. High soy intake has also been related to weight loss in women with polycystic ovary syndrome (PCOS), which may improve health and disease results [68]. The results of the above studies show that food ingredients may not always have a direct impact on fertility and their indirect impact is equally important. However, more studies concerning soy intake on fertility and women's health are needed. Many studies suggest that high soy intake may cause interference with ovarian function because of its high phytoestrogen content [69,70]. More studies are needed to determine the appropriate intake of soy of women trying to conceive, since an excessive intake of soy may not be safe.

### *4.6. Folate*

The Center for Disease Control and Prevention (CDC) in the United States, recommends that healthy women with a low risk of birth defects should supplement 400 μg a day of folic acid at least 12 weeks before conception and early pregnancy to avoid neural tube defects [71]. However, it is unknown whether folate intake is related to female fertility. Recent studies have turned to methylenetetrahydrofolate reductase (MTHFR) gene mutations as the cause of recurrent miscarriages [72]. It seems that supplementation of vitamins B6, B12 and supraphysiologic methylfolate could help women with MTHFR gene mutations to conceive [73]. A high intake of folate and B12 was associated with an increased birth rate in women undergoing ART [59]. A high folate-to-homocysteine ratio was related to a lower risk of anovulation in regularly menstruating females [60]. Interesting retrospective studies were also conducted on supplemental folate intake and pollutant exposure among women undergoing ART. The subjects with high pollutant exposure had a lower rate of live birth [35] or implantation ratio [34]; however, folate supplementation positively modified these results in both studies. These results agree with the animal studies mentioned previously [26,27].

### *4.7. Vitamin D*

During the current study, a number of studies were found which analysed the effect of vitamin D [8–11,36,49,74–78]. Nine of them showed a positive, statistically significant association with female fertility [8,9,11,36,49,74,76–78]. However, since only possible interactions with food compounds were searched for, and vitamin D is mostly formed under sun exposure, two of them concerning vitamin D dietary intake were included in the review. Summing up the issues of vitamin D and fertility, it should be emphasised that food products fortified with vitamin D could be advisable [49] for some populations, not only because of the fertility support but also overall health [79].

### *4.8. Antioxidants*

The search also resulted in study findings involved in other single micronutrients. In a Russian study concerning serum metal concentration in the blood of women, it was found that females with infertility and miscarriage had a lower concentration of copper than pregnant women [55]. Moreover, there was no difference between the copper levels of pregnant women and a healthy control group, which supports the hypothesis that copper may play a role in female fertility.

A diet rich in antioxidants, such as the Mediterranean dietary pattern could improve fertility, although the only recent study concerning the intake of antioxidants did not support this hypothesis. Moreover, the intake of foods with a high concentration of β-carotene, lutein and zeaxanthin was

inversely associated with live birth rates [37]. The reasoning of these results may be variable, starting from the accuracy of the food frequency questionnaire used in the study to the dietary pattern which leads to these results. The most important is the fact that supplementing antioxidants as fertility support or for other health conditions could be dangerous and should not be advised to the patients.

### *4.9. Other Food Compounds*

Most of the studies found in the systematic search were qualitative and retrospective, although four prospective randomised trials published between 2015 and 2019 were also included in the review (Table 2). The studies concerned coenzyme Q10 [48], melatonin [58] and multiple nutrient supplementation of vitamin E and D [52] as well as numerous micronutrient or only folate supplement [47]. All of the included study interventions had a positive effect on the pregnancy rate and should be further studied to support fertility treatment nutritionally.

### **5. Conclusions**

In conclusion, it should be noted that reproductive performance is influenced by food and type of nutrition. The findings of the current study suggest that the importance of food production. In particular, the availability and intake of pro-healthy food compounds is a significant factor supporting fertility in females. Women planning pregnancy should especially ensure an intake of fruits and vegetables, folate and vitamin D. A high intake of sweetened beverages and trans-fatty acids appears to decrease the chance of pregnancy and live birth rate. Soy food intake needs to be further analysed because of its endocrine-disruptive properties. Environmental pollution influences fertility around the world, and the appropriate intake of bioactive food compounds could diminish this effect. However, the excessive intake of specific micronutrients, such as antioxidants, may decrease fertility. More randomised prospective studies are needed to analyse the impact of micronutrients on fertility, taking into consideration the patient environment and environmental pollution.

**Author Contributions:** Conceptualisation, M.C.-M. and A.B.-D.; methodology, M.C.-M., M.K. and A.B.-D.; investigation, A.B.-D. and E.K.; resources, M.C.-M.; data curation, A.B.-D. and E.K.; writing—original draft preparation, A.B.-D.; writing—review and editing, A.B.-D., E.K., M.K., M.C.-M.; supervision, M.C.-M.; project administration, M.C.-M.; funding acquisition, M.C.-M.; All authors have read and agreed to the published version of the manuscript.

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

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

### **References**


© 2020 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 (http://creativecommons.org/licenses/by/4.0/).
