An automated cervical cancer diagnosis using genetic algorithm and CANFIS approaches

Abstract

BACKGROUND:

Cervical malignancy is considered among the most perilous cancers affecting women in numerous East African and South Asian nations, both in terms of its prevalence and fatality rates.

OBJECTIVE:

This research aims to propose an efficient automated system for the segmentation of cancerous regions in cervical images.

METHODS:

The proposed techniques encompass preprocessing, feature extraction with an optimized feature set, classification, and segmentation. The original cervical image undergoes smoothing using the Gaussian Filter technique, followed by the extraction of Local Binary Pattern (LBP) and Grey Level Co-occurrence Matrix (GLCM) features from the enhanced cervical images. LBP features capture pixel relationships within a mask window, while GLCM features quantify energy metrics across all pixels in the images. These features serve to distinguish normal cervical images from abnormal ones. The extracted features are optimized using Genetic Algorithm (GA) as an optimization method, and the optimized sets of features are classified using the Co-Active Adaptive Neuro-Fuzzy Inference System (CANFIS) classification method. Subsequently, a morphological segmentation technique is employed to categorize irregular cervical images, identifying and segmenting malignant regions within them.

RESULTS:

The proposed approach achieved a sensitivity of 99.09%, specificity of 99.39%, and accuracy of 99.36%.

CONCLUSION:

The proposed approach demonstrated superior performance compared to state-of-the-art techniques, and the results have been validated by expert radiologists.

Keywords

Cervical cancer gaussian filter features optimization genetic algorithm classification

1. Introduction

Cervical malignancy stands out as one of the most perilous cancers affecting women in various East African and South Asian nations, both in terms of its prevalence and fatality rates. In 2008 alone, around 530,000 women globally received diagnoses of invasive cervical cancer, with 275,000 succumbing to this disease. Breast and cervical cancers disproportionately impact women [1]. The onset of cervical cancer, characterized by malignant cells in the cervix region, often leads to bleeding in later stages, resulting in sudden fatalities [2].

Figure 1a depicts a typical cervical section in females without malignancy, while Fig. 1b illustrates the abnormal malignancy region in a woman afflicted by cervical cancer. The primary culprit for cervical malignancy in females is the presence of the Human Papilloma Virus (HPV) in the cervical section. According to the World Health Organization (WHO), over eighty percent of cervical malignancy cases are on the rise in emerging nations, primarily due to a lack of information and awareness about the illness [3]. The shortage of clinician or surgeon resources in developing nations emphasizes the critical need for examining cervical malignancy using computerized machine learning methods [4]. The ongoing research for malignancy section identification in cervical images serves various clinical and health settings, aiding physicians in the automated detection of malignancy sections [5].

In light of these challenges, this research proposes a machine learning-based intuitive detection and categorization method for cervical malignancy recognition and fragmentation. While numerous machine learning algorithms have been employed for cancer image classifications in the past, the CANFIS method emerges as particularly effective for cervical cancer detection systems, owing to its proficient supervised internal layers. The CANFIS classifier, a modification of the ANFIS classifier, enhances fuzzy rules to yield superior classification results. Additionally, the Genetic Algorithm (GA) optimization algorithm operates on external feature metrics to refine computed feature values in cervical images. In this paper, the GA method optimizes the computed feature set, which is subsequently classified using the CANFIS classifier.

Figure 1.

Cervical images (a) Normal case (b) Abnormal case.

Section 2 of this research work describes the present approaches for cervical malignance finding and fragmentation. Sections 3 and 4 create a novel strategy for detecting and classifying cervical malignance. Section 5 discusses simulation outcomes. Section 6 designates this research work conclusion, and Section 7 discusses future improvements.

2. Literature review

Pan et al. [6] employed a multimodal system or methodology for screening cervical cancer using an MRI staging model. The authors discussed various non-linear factors to distinguish between normal and abnormal images. Ghaffar et al. [7] utilized various optimization techniques and explored their implications with elderly disabled systems. Lin et al. proposed a cervical healthy computing system for analyzing various healthcare parameters to predict or screen cervical abnormalities. Umesh Kumar Lilhore et al. [8] implemented machine modeling algorithms for the detection of cervical cancer using cervigram images in their work. The authors used the Boruta analysis algorithm to predict features from cervical images. These features were combined and classified through a hybrid machine model that employed the combination behavior of the Support Vector Machine (SVM) and the Random Forest (RF) to improve the classification rate of the cervical cancer detection system. Chitra et al. [9] utilized the fuzzy local information c-means clustering algorithm to cluster abnormal pixels in cervical images. The authors applied the Sooty Tern Optimization framework model to optimize the region of pixels belonging to the cancer region in their work. Furthermore, a CNN-Long Short Term Memory (LSTM) model was employed to produce classification results. Pallavi et al. [10] analyzed Pap smear cell images to predict cervical cancer using various supervised learning techniques. The hyperbolic tangent functional logic was employed in the supervised classifier. This work utilized an SVM classifier to predict non-linear features and produce results. Vijayanand Sellamuthu Palanisamy et al. [11] designed a cervical cancer detection algorithm using residual neural networks. The activation function was optimized using the Adam optimization algorithm in their work. Ontor et al. [12] proposed a YOLO deep learning model for the detection of cervical cancer. This method used a minimized set of layers to optimize the computational complexity of the cancer detection algorithm. The authors achieved a 96.1% classification rate using the proposed model. Wang et al. [13] proposed a feature computational modeling method for detecting abnormalities in medical images. Kamal et al. [14] developed a system between the human brain and a computer system to acquire brain activity signals, which were then used by soft computing algorithms to predict disease status. Sahli et al. [15] computed a geometrical feature set from medical images, and these computed geometrical features were used by various classification algorithms.

Alyafeai Zaid et al. [16] employed a deep learning-based classification to detect and segment malignancy sections in cervical images. The authors utilized the Guanacase Dataset 2005 and achieved a standard sensitivity (SE) of 97.2%, standard specificity (SP) of 98.3%, and standard accuracy (AC) of 98.8% compared to ground truth images. Thiyaneswaran et al. [17] utilized the k-means clustering method to detect and segment malignancy sections in surface images. The authors achieved a standard accuracy (AC) of 90.0% when using an unfolded admittance database. Wasswa William et al. [18] proposed a pap smear investigation mechanism for cervical malignancy detection from pap smear images. They achieved a standard sensitivity (SE) of 99.28%, standard specificity (SP) of 97.4%, and standard accuracy (AC) of 98.8%. Sachan et al. utilized the random forest classification strategy on cervical images. They achieved 90.5% standard sensitivity (SE), 93.2% standard specificity (SP), and 92.9% standard accuracy (AC) from the unfolded admittance database. Karthiga Jaya et al. [19] detected and segmented malignancy sections in cervical images using the ANFIS categorization strategy. The authors utilized the Guanacase Database 2005 and achieved a standard sensitivity (SE) of 97.42%, standard specificity (SP) of 98.3%, and standard accuracy (AC) of 99.12% compared to ground truth images.

Kumarganesh et al. [20] proposed an ANFIS classifier strategy for tumor categorization from original images, achieving a categorization standard accuracy (AC) of 96.6%. Kumarganesh et al. [21] proposed using an ANFIS categorizer strategy to detect tumors in original images. They achieved a standard sensitivity (SE) of 93.07%, standard specificity (SP) of 98.79%, and standard accuracy (AC) of 97.63% for cancer categorization. Teck et al. [22] employed the contourlet transformation strategy in conjunction with the Gabor orientation strategy to transform the original cervical images. The converted coefficients were combined by the second order arithmetical component specifications, and the resulting component set was well qualified and classified using the SVM categorizer. The authors achieved a standard categorization accuracy (AC) of 81% in detecting malignancy portions in original cervical images.

Yuexiang et al. [23] introduced a direct strategy for identifying malignancy portions in cervical images used in the input test. The original cervical images were employed to determine first and second-order arithmetical components, which were then fine-tuned and classified using the linear categorization strategy. Finally, to detect cancer regions in the categorized outcome images, a morphological fragmentation strategy was applied. Fei et al. [24] proposed a procedure to distinguish abnormal regions in the cervix areas of female patients. The authors used biopsy images to scan the cervical portions of the patients using a Computer-Aided Detection (CAD) strategy. This computer-aided strategy, employing various soft computing techniques, enhanced the categorization precision of the cervical malignancy exploration scheme. In comparison to ground truth images, the authors achieved a standard sensitivity (SE) of 86% and a standard tumor sectionalization accuracy (AC) of 91%.

Slowik et al. [25] designed an optimal strategy for deciding and separating the Pap smear cells of patients. Based on their morphological attributes, the authors compared and evaluated the fragmented nucleus section of cervical images using the physician fragmented nucleus portions, achieving a standard categorization accuracy (AC) of 93%. For fragmenting the cancer portion in cervical images, Elayaraja et al. [26] developed a GA-based CANFIS categorization mechanism that achieved 99.37% standard sensitivity (SE), 98.9% standard specificity (SP), and 99.21% standard accuracy (AC). Senthil Kumar et al. [27] improved accuracy by extracting performance assessment measures using an image thickening and background thinning technique. Thiyaneswaran et al. [28] calculated that an AlexNet with an ADAM solver achieved a system accuracy of 98.21%. Prakash et al. proposed a hybrid CNN method and RIDGELET transform for the detection and classification of Meningioma brain tumors, attaining a classification accuracy of 99.7% in the BRATS 2019 dataset and 99.36% in the Nanfang dataset [29].

3. Materials and methods

3.1 Materials

The Guanacaste database was utilized to acquire the original cervical images for this study. This dataset includes a substantial number of original cervical images paired with corresponding ground truth images that have been validated by a specialist physician. In this research work, the images from this database are categorically assigned into various severity levels.

3.2 Methodology

This research introduces an effective approach to identifying and isolating malignant regions in cervical images at their source. The method utilizes a Genetic Algorithm (GA)-based optimization strategy. Initially, a Gaussian filter is applied to enhance the pixels in the cancer-affected areas of cervical assessment images, ensuring accurate classification. Subsequently, specific attributes are removed from the filtered images. These discarded parameters are then refined using GA. The improved attributes are employed for classification through the CANFIS categorization method. Finally, morphological operations are applied to segment the malignant regions in abnormal cervical images [30]. The proposed strategy for detecting and categorizing these malignant areas in cervical images is illustrated in Fig. 2.

Figure 2.

Proposed cervical cancer detection and segmentation method.

3.2.1 Preprocessing

Image smoothing is one of the preprocessing techniques employed in this research, achieved by utilizing the Gaussian filter [31] on the cervical images. The Gaussian kernel is defined as:

$\displaystyle G(x)=\frac{1}{\sqrt{2\pi\sigma^{2}}}{e}^{-\left({\frac{x^{2}}{2% \sigma^{2}}}\right)}$ (1)

Where ‘ $x$ ’ represents the pixel value, and ‘ $\sigma$ ’ stands for the standard deviation, assumed to have a mean value of zero. The two-dimensional Gaussian kernel is derived by combining the properties of two one-dimensional Gaussian kernels. This approach is applied to cervical images to obtain smoother and flatter regions within them.

3.2.2 Feature extractions

The process of capturing or eliminating attributes from the original cervical images is referred to as feature extraction. This operation is conducted to facilitate the classification of the original cervical images. By extracting features from various regions of the image, the original cervical image is categorized. The resulting cervical images are classified into one of two categories: normal or abnormal. The following features are extracted from the cervical images that have undergone a Gaussian filter in order to categorize them.

3.2.2.1 LBP

This non-parametric descriptor generates a unique pattern based on the values of its neighboring pixels. It encodes all the pixels in a Gabor-transformed image into a distinct attribute parameter, highly useful for pattern recognition in various computer vision applications. In the process of computing the Local Binary Pattern (LBP), as described by Prasad et al. in 2023 [32], a 3x3 overlapping window is applied over the Gabor-transformed image. For each center pixel within this window, the value is computed by subtracting it from the surrounding pixels within the window. A binary value of ‘0’ is assigned if the subtraction result is negative, and ‘1’ is assigned if the subtraction result is positive or equal to ‘0’. The LBP feature of the center pixel in the 3x3 window is calculated using the following equation.

$\displaystyle\textit{LBP}({P_{c}})=\sum\limits_{{i}=0}^{N}{S(P_{c}-P_{i})}*{2}$ (2)

Whereas $P_{c}$ is the center pixel in the 3x3 window, and $P_{i}$ is the neighboring picture element of the middle picture elements. Moreover, $N$ is the number of surrounding pixels neighboring the center pixel. ‘ $S$ ’ denotes the scaling function of the LBP operator, and the following equations represent it:

$\displaystyle S(k)=\left\{{\begin{array}[]{l}1,\text{if }k\geqslant 0\\ 0,\text{if }k<0\\ \end{array}}\right.$ (3)

The LBP features are primarily used to detect variations in non-smoothed regions in cervical images. The LBP-extracted image of the Gabor-transformed cervical picture is depicted in Fig. 3.

Figure 3.

LBP extracted cervical picture.

3.2.2.2 GLCM features

GLCM (Gray-Level Co-occurrence Matrix) attributes, as described by Lin et al. in 2022 [33], are texture-related features that depend on the number of pixel pairs with specific concentration values. These attributes are established in a specific spatial arrangement and are used to analyze the relationships between the distributions of pixel values in a cervical image. For the computation of GLCM attributes, one of the directional distributions at angles 0^∘, 45^∘, 90^∘, and 180^∘ can be utilized. In this context, the rows and columns of the GLCM matrix are denoted by ‘i’ and ‘j,’ respectively. ‘I(i, j)’ represents a cervical image that has undergone Gaussian filtering. The GLCM feature set includes the following characteristics: homogeneity, contrast, energy, correlation coefficient, variance, maximum probability, and inverse dissimilarity. These GLCM attributes are employed to distinguish between abnormal and normal regions in the cervical image and are extracted from the Gabor-transformed cervical image to aid in this differentiation process.

The standards for GLCM attributes for the regular cervical picture, which is free of lesions, and the aberrant cervical picture, which has cancer patterns, are listed in Table 1. The equations depicted in Table 1 are used to compute the GLCM features, and these values vary with respect to normal and abnormal cervical images.

Table 1

GLCM features for normal and abnormal cervical images

GLCM features	Computation expression
Entropy	$\sum\limits_{i}^{M}\sum\limits_{j}^{N}{P[{i,j}]}\log P[{i,j}]$
Homogeneity	$\sum\limits_{i=1}^{M}\sum\limits_{j=1}^{N}{\frac{P[{i,j}]}{1+\|{i-j\|}}}$
Contrast	$\sum\limits_{i}^{M}\sum\limits_{j}^{N}{({i-j})}^{2}P[{i,j}]$
Energy	$\sum\limits_{i}^{M}\sum\limits_{j}^{N}P^{2}[{i,j}]$
Correlation coefficient	$\sum\limits_{i=1}^{M}\sum\limits_{j=1}^{N}{\frac{({i-\mu})({j-\mu})P[{i,j}]}{% \sigma^{2}}}$
Variance	$\frac{{1}}{{2}}\sum\limits_{i=1}^{M}\sum\limits_{j=1}^{N}{({({i-\mu})^{2}P[{i,% j}]+({j-\mu})^{2}P[{i,j}]})}$
Maximum probability	$\mathop{\text{max}}\limits_{{i,j}}^{{M,N}}P[{i,j}]$
Inverse difference moment	$\sum\limits_{i=1}^{M}\sum\limits_{j=1}^{N}{\frac{P[{i,j}]}{\|{i-j\|}^{k}}},i\neq% {j}$

The non-linearity performances of the retrieved attributes are developed using GA. In this proposed study, the dimensions of the attributes are reduced using GA, and the optimized set of attributes is then fed into the categorization procedure to classify images of the cervical region.

3.2.3 Features optimization

The feature set extracted from the original cervical image needs to be organized into a suitable matrix-vector format. To streamline and reduce the length of this feature set, it is crucial to apply an optimized algorithm. With the assistance of a Genetic Algorithm (GA), the nonlinearity characteristics of the extracted feature set are fine-tuned. This optimization process aims to enhance the classification performance of the proposed system by focusing on the most relevant and optimized features. In this paper, the application of GA allows for the reduction of the dimensions of the extracted attributes. These optimized attribute sets are subsequently utilized as inputs for the cervical image classification method.

The GA (Genetic Algorithm) algorithm is outlined in the following steps:

Step 1: Initiate the process with a randomly selected population, where each chromosome comprises a variety of genes. To form the initial population, replicate it 70 times. Step 2: Initialize the weights for all genes on the chromosomes with randomly assigned starting values. Step 3: Apply the crossover function to the genes within the chromosome along with their associated weight values. The crossover probability is set to 0.5 to achieve the best optimization in this context. Step 4: After the crossover process, apply the mutation function to the resulting chromosomes. A mutation probability of 0.2 is chosen to achieve the best optimization. Step 5: Once the mutation process is complete, compute the Euclidean distance between each gene and its corresponding chromosome. Select the chromosome with the smallest Euclidean distance. Step 6: Repeat steps 2 through 6 for each feature in the attribute set matrix until all features have been processed and optimized.

This algorithm outlines the steps involved in optimizing a set of features using a genetic algorithm for a specific task or application.

The possibility cost of spilling and merge processes stabilizes at the conclusion of the 1st iteration, and the resulting mathematical formulas can be used to calculate them.

The formula for calculating the possibility of a split process is

$\displaystyle\Pr ob(\textit{split})=0.5-\textit{Minimum}\left({\frac{N}{50},0.% 5}\right)$ (4)

Whereas, the populace volume N in this work is initially fixed to 70.

The formula for calculating the possibility of a merging process is

$\displaystyle\Pr ob(\textit{merge})=0.5-\textit{Minimum}\left({\frac{N}{50},0.% 2}\right)$ (5)

3.2.4 Classifications

ANFIS (Adaptive Neuro-Fuzzy Inference System), as extended by Senthil Kumar et al. in 2020 [34], has been expanded to coactive ANFIS, designed to handle multiple outputs and can be considered as a generalization of ANFIS. A modified classification architecture derived from ANFIS is known as CANFIS (Coactive Neuro-Fuzzy Inference System).

The fundamental component of CANFIS is the fuzzy neuron, responsible for assigning membership values to the input. Two commonly used types of membership functions are the universal Bell and Gaussian membership functions. The membership function for a fuzzy set over the input universe X is defined as $\mu$ A: X $\to$ [0, 1], where it maps every element of X to a value between 0 and 1. This value quantifies the degree of membership of an element in X to the fuzzy set A, often referred to as the membership value or degree of membership. In addition to the fuzzy neuron, a normalization function is included in the network to scale the output within the range of 0 to 1. CANFIS is a versatile framework used for various tasks, including classification and inference, by effectively combining fuzzy logic and neural network principles.

The core element of CANFIS is the fuzzy neuron, which assigns membership values to the input. Two commonly employed types of membership functions are the universal Bell and Gaussian membership functions. The membership function for a fuzzy set over the input universe X is defined as $\mu$ A: X $\to$ [0, 1], mapping each element of X to a value between 0 and 1. This value quantifies the degree of membership of an element in X to the fuzzy set A, commonly referred to as the membership value or degree of membership. In addition to the fuzzy neuron, a normalization function is incorporated into the network to scale the output within the range of 0 to 1. CANFIS serves as a versatile framework utilized for various tasks, including classification and inference, effectively combining principles from fuzzy logic and neural networks.

Figure 4.

CANFIS categorization structure of the proposed method.

In Fig. 5, each layer plays a specific role in predicting cancer, described as follows:

Layer 1: Each node in Layer 1 assesses the extent to which the given input belongs to one of the fuzzy sets (A, B, C, or D) by computing the membership grade for each fuzzy set. Three membership functions define these fuzzy sets.

Layer 2: The combination of all the outputs from Layer 1 serves as the input to Layer 2. Layer 2 consists of two components. The lower component represents the modular network, calculating the total firing strength for each output. The higher component utilizes the membership functions assigned to each of the inputs. This architecture integrates fuzzy logic principles with neural network elements to predict cancer based on input data. The membership grades from Layer 1 and the modular network’s computations in Layer 3 contribute to making accurate predictions regarding cancer-related outcomes.

The final output of the network is computed in the 4th layer, which calculates the weight regularization of the results from the two components of the 3rd layer. In this design, there is no need for a separate Layer 5, as the functions it would typically perform are handled within Layer 4 itself. This streamlined architecture simplifies the network while achieving the desired outcome.

Figure 5.

CANFIS categorization architecture for predicting cancer.

In the CANFIS categorization structure, the X node receives input features from the original cervical image, while the Y node receives the trained patterns, encompassing both normal and anomalous examples. Based on the learned patterns, this structure generates an output of either 0 or 1, where binary values 0 and 1 represent the categories of normal and abnormal, respectively.

The inputs for this categorizer are denoted as X and Y, and the output can be either a training pattern or a categorized pattern, depending on the mode of this classifier. The CANFIS categorization process for each layer is described by the following equations, detailing how the network processes input data and generates the final categorization result.

The symbols for Layer 1 are,

$\displaystyle G_{1,i}=\vartheta_{Ai}(x)\text{ for }i=1,2,\ldots$ (6) $\displaystyle G_{1,i}=\vartheta_{Bi}(x)\text{ for }i=3,4,\ldots$ (7)

Where $x$ and $y$ are the computed features with respect to normal and abnormal cervical image respectively during the training of the classifier.

Layer 2 is represented by,

$\displaystyle\vartheta_{A}(x)=\frac{1}{1+\left|{\frac{x-c_{i}}{a_{i}}}\right|^% {2b_{i}}}$ (8)

Where, $a_{i}$ and $c_{i}$ are the arbitrary constant and cumulative variable and $b_{i}$ is the enhancing parameter.

$\displaystyle G_{2,i}=w_{i}=\vartheta_{A_{i}}(x)\cdot\vartheta_{B_{i}}(y),i=1,% 2,\ldots$ (9)

Layer 3 is represented by,

$\displaystyle G_{3,i}={w}_{i}=\frac{{w}_{i}}{{w}_{1}+{w}_{2}}$ (10)

where, $w_{1}$ and $w_{2}$ are the weights of the nodes in CANFIS architecture.

The symbols for Layer 4 are,

$\displaystyle G_{4,i}=\overline{w_{i}}f_{i}=\overline{w_{i}}({p_{i}x+q_{i}y+r_% {i}})$ (11)

Where as, $p_{i}$ , $q_{i}$ , and $r_{i}$ are the layer metric parameters.

Layer 5 is represented by,

$\displaystyle G_{5,i}=\sum_{i}{\overline{w_{i}}f_{i}}=\frac{\sum_{i}{{w}_{i}f_% {i}}}{\sum_{i}{{w}_{i}}}$ (12)

Whereas, $f_{i}$ is the functional output of the layer and w is the internal layer weight.

In the CANFIS categorization process, the cervical image is classified as “normal” (benign) when the response from Layer 5 is low, and it is classified as “abnormal” (malignant) when the response from Layer 5 is high. Using CANFIS categorization, the features are effectively categorized into normal or abnormal classes. Following the categorization, further morphological operations are applied to the classified abnormal cervical image to segment the malignancy section. This segmentation involves the use of two fundamental morphological operations:

Erosion Operation: This operation is applied to the irregular cervical image to obtain an eroded image.

Dilation Operation: The dilation operation is applied to the irregular cervical image to obtain an expanded or dilated image.

To identify and isolate the cancerous region within the cervical image, the eroded image is subtracted from the dilated image. This process helps to highlight and separate the cancerous area from the surrounding tissues, aiding in the accurate localization and analysis of malignant regions in the cervical image.

4. Results and discussion

The cervical images utilized in this research were obtained from the Guanacaste database [35]. This dataset contains cervical images with manually segmented cancer regions, marked by experienced radiologists, providing ground truth annotations. These ground truth cervical images play a crucial role in assessing the performance of the proposed system using various performance metrics. The dataset comprises a total of 295 cervical images, further divided into training and testing sets. The dataset includes 120 cervical images with normal cases and 175 cervical images with abnormal cases. The proposed cervical malignancy detection and segmentation approach were tested on these images. The experimental results obtained on the 295 cervical images from this database demonstrate the effectiveness of the proposed method. Additionally, the same experimental results can be extended to other sets of cervical images obtained from real-time clinical datasets or similar cervical image databases, indicating the generalizability of the approach to various datasets and clinical scenarios. The projected strategy is simulated in this work using MATLAB R2015 and incorporates the GA strategy along with the CANFIS categorization approach.

The categorization rate, which is described in the following mathematical formula for both normal and abnormal cases, is used to analyze the effectiveness of the suggested cervical malignance finding method.

$\displaystyle\textit{Classification }\text{rate for normal case}=\frac{\text{% Number of images correctly detected (normal case)}}{\text{Total Number of images (normal case)}}$ (13) $\displaystyle\textit{Classification }\text{rate for abnormal case}=\frac{% \begin{array}[]{c}\text{Number of images correctly detected}\\ \text{(abnormal case)}\end{array}}{\text{Total Number of images (abnormal case% )}}$ (14)

Table 2 presents an evaluation of the anticipated strategy’s performance based on the categorization rate for the average case. The projected technique GA $+$ CANFIS correctly classifies 119 images out of 120 images for normal cases, achieving a classification rate of 99.1%. The method specified by Alyafeai Zaid et al. [16] classifies 117 images out of 120 images for normal cases, resulting in a classification rate of 97.5%. The approach outlined by Wasswa William et al. [18] classifies 115 images out of 120 images for normal cases, achieving a classification rate of 95.8%. The method described by Karthiga Jaya et al. [19] classifies 110 images out of 120 images for normal cases, achieving a classification rate of 91.6%.

Table 2 clearly demonstrates that the proposed strategy attains a significantly higher classification rate for normal cases compared to other conventional systems. The performance of the proposed approach is further illustrated in Fig. 6, where it is graphically compared based on the classification rate for the average case.

Table 2

Performance comparisons of projected strategy in terms of categorization rate for normal case

Methodology	Number of images for normal case	Number of images correctly classified	Classification rate (%)
GA $+$ CANFIS (proposed system)	120	119	99.1
Alyafeai Zaid et al. [16]	120	117	97.5
Wasswa William et al. [18]	120	115	95.8
Karthiga Jaya et al. [19]	120	110	91.6

Table 3 provides a comprehensive analysis of the proposed technique’s performance in terms of classifying abnormal cases. The GA $+$ CANFIS strategy demonstrates remarkable performance by correctly classifying 174 out of 175 images for the abnormal cases, achieving an impressive classification rate of 99.4%. In comparison, the conventional approach presented by Alyafeai Zaid et al. in 2019 [16] correctly classifies 172 out of 175 images for the abnormal cases, resulting in a classification rate of 98.2%. The method proposed by Wasswa William et al. in 2019 [18] accurately classifies 170 out of 175 images, achieving a classification rate of 97.1%. Finally, the approach presented by Karthiga Jaya et al. in 2018 [19] correctly classifies 169 out of 175 images, resulting in a classification rate of 96.5%. These results clearly underscore the superior performance of the GA $+$ CANFIS strategy in accurately identifying and classifying abnormal cases when compared to other existing methods.

The proposed strategy, as demonstrated in Table 3, clearly outperforms other conventional methods, achieving a significantly higher classification rate for the abnormal cases.

Table 3

Performance comparisons of projected strategy in terms of classification rate for abnormal case

Methodology	Number of images for abnormal case	Number of images correctly classified	Classification rate (%)
GA $+$ CANFIS (proposed system)	175	174	99.4
Alyafeai Zaid et al. [16]	175	172	98.2
Wasswa William et al. [18]	175	170	97.1
Karthiga Jaya et al. [19]	175	169	96.5

Figure 6.

Graphical comparisons of proposed methods in terms of categorization rate for standard case.

Figure 7 shows the graphical comparisons of the projected strategy in terms of the categorization rate for the abnormal case.

With the assistance of performance assessment specifications, the effectiveness of the suggested cervical malignancy classification and separation approach is examined. The metrics include sensitivity (SE), specificity (SP), accuracy (AC), Positive Predictive Value (PPV), and Negative Predictive Value (NPV), as defined in the previous chapter.

The proposed GA with CANFIS categorization method for the detection and categorization of cervical malignancy, as stated in this research study, achieves 99.09% standard sensitivity, 99.39% standard specificity, and 99.36% standard accuracy, 99.04% standard PPV, and 90.96% standard NPV, as represented in Table 4.

Table 4

Performance estimation of projected cervical malignance finding scheme

Image sequences	Sensitivity	Specificity	Accuracy	PPV	NPV
1	98.5	99.4	99.2	99.6	92.9
2	98.4	98.8	98.8	98.8	88.3
3	99.5	99.8	99.4	99.1	92.1
4	98.9	99.7	99.8	99.3	90.5
5	99.3	99.8	99.8	99.5	90.6
6	98.9	99.3	99.4	99.2	89.6
7	99.6	98.9	98.8	98.7	92.7
8	99.5	99.1	99.4	99.1	90.6
9	99.4	99.3	99.6	98.4	92.7
10	98.9	99.8	99.4	98.7	89.6
Average	99.09	99.39	99.36	99.04	90.96

Figure 7.

Graphical comparisons of proposed strategy in terms of categorization rate for abnormal case.

Figure 8 demonstrates the graphical analysis of performance evaluation for the planned cervical cancer detection system. Table 5 provides performance assessments of the projected strategy compared to predictable cervical malignancy detection technologies.

The GA with CANFIS categorization technology specified in this work achieved 99.09% standard SE (sensitivity), 99.39% standard SP (specificity), and 99.36% standard AC (accuracy). In comparison, the conservative process specified by Alyafeai Zaid et al. [16] accomplished 97.2% standard SE (sensitivity), 98.3% standard SP (specificity), and 98.8% standard AC (accuracy) in relation to images that represent the real world.

The conventional method Wasswa William et al. [18] achieved 99.28% standard SE (sensitivity), 97.4% standard SP (specificity), and 98.8% standard AC (accuracy). Karthiga Jaya et al. [19] accomplished 97.42% standard SE (sensitivity), 98.3% standard SP (specificity), and 99.12% standard AC (accuracy) in relation to images that represent the real world.

Table 5

Performance comparisons of projected strategy with conservative cervical malignance finding technologies

Methodology	Sensitivity (%)	Specificity (%)	Accuracy (%)
GA based CANFIS classification (proposed system)	99.09	99.39	99.36
Alyafeai Zaid et al. [16]	97.2	98.3	98.8
Wasswa William et al. [18]	99.28	97.4	98.8
Karthiga Jaya et al. [19]	97.42	98.3	99.12

Figure 8.

Graphical analysis of Performance assessment for projected cervical malignance finding scheme.

Figure 9 displays the graphic illustrations of performance assessments of proposed the process with conservative cervical malignance finding technologies.

Figure 9.

Graphical illustrations of performance comparisons of proposed method with conventional cervical cancer detection methodologies.

5. Conclusions and future scope

This research introduces an enhanced categorization process utilizing the CANFIS categorization procedure to detect and classify malignancies in cervical images. The methodology involves initial image enhancement through the Gaussian Filter technique, followed by attribute extraction using LBP and GLCM from the enhanced cervical images. These extracted attributes undergo optimization using the Genetic Algorithm (GA), and the optimized attribute sets are then classified using the CANFIS categorization method. Subsequently, a morphological segmentation technique is applied to the categorized abnormal cervical images for the detection and segmentation of cancerous regions. The GA-based CANFIS categorization methodology achieves impressive standard sensitivity (SE) of 99.09%, specificity (SP) of 99.39%, and accuracy (AC) of 99.36%. The GA $+$ CANFIS classification system accurately classifies 119 out of 120 images, resulting in a 99.1% classification rate. Furthermore, the GA $+$ CANFIS classification method correctly classifies 174 out of 175 images, achieving a remarkable 99.4% classification rate. However, it is essential to acknowledge certain limitations in this study. While the methodologies outlined here lead to more accurate cervical cancer detection and segmentation, they lack further analysis of the severity level of the segmented cancer regions. This is a notable limitation, as many radiologists require information on the severity of detected malignancies. Additionally, the proposed cervical cancer detection system relies on machine learning methods like CANFIS, which demand a substantial number of cervical images during the training stage, consuming more time for cervical cancer screening.

Looking ahead, future approaches will likely revolutionize cervical imaging detection by classifying cervical images into severity categories, such as “Early,” “Advanced,” or “Surgery,” enhancing medical decision-making. Streamlining pre-processing, segmentation, and categorization will significantly reduce the time required for each stage, offering faster results for healthcare providers. The final step involves developing a cervical image analysis system utilizing emerging hybrid technologies like Field Programmable Gate Array (FPGA), Internet of Things (IoT), and Very Large-Scale Integration (VLSI). This innovative approach will empower medical practitioners to deliver treatment recommendations more efficiently. The completion and evaluation of an FPGA-based hardware implementation, alongside an IoT-based real-time patient monitoring system, mark the next steps in advancing cervical cancer detection technology.

Data availability statement

The data used for this research is a publicly archived dataset available from: http://www.cse.lehigh.edu/idealab/cervitor/downloads.html.

Funding

No funding was received for this research.

Footnotes

Conflict of interest

The authors declare that there is no conflict of interest.

References

Chen

Xue

. Computer-aided cervical cancer diagnosis using time-lapsed colposcopic images. IEEE Transactions on Medical Imaging. 2020; 39(11): 3403-3415.

Guo

Banerjee

Stanley

. Nuclei-based features for uterine cervical cancer histology image analysis with fusion-based classification. IEEE Journal of Biomedical and Health Informatics. 2016; 20(6): 1595-1607.

Chen

Yang

, et al. LncRNA GAS5 regulates redox balance and dysregulates the cell cycle and apoptosis in malignant melanoma cells. J Cancer Res Clin Oncol. 2019; 145: 637-652.

Curry

Krist

. Screening for cervical cancer US preventive services task force recommendation statement. JAMA. 2018; 320(7): 674-686.

American Cancer Society (ACS) What is cervical Cancer? 2011. Available at: http://www.Cancer.org/Cancer/CervicalCancer/Detailed-Guide/cervical-Cancer-what-is-cervical-Cancer.

Liang

, et al. Application of 3.0 T Multimodal MRI in FIGO Staging of Cervical Cancer. Technology and Health Care. 2023; Pre-press(Pre-press): 1-8.

Ghaffar

Dehghani-Sanij

Xie

. Actuation system modelling and design optimization for an assistive exoskeleton for disabled and elderly with series and parallel elasticity. Technology and Health Care. 2023; 31(4): 1129-1151.

Lilhore

Poongodi

Kaur

Simaiya

Algarni

Elmannai

Vijayakumar

Tunze

Hamdi

. Hybrid Model for Detection of Cervical Cancer Using Causal Analysis and Machine Learning Techniques. Computational and Mathematical Methods in Medicine, vol. 2022, Article ID 4688327, 17 pages, 2022.

Chitra

Kumar

. Early cervical cancer diagnosis using Sooty tern-optimized CNN-LSTM classifier. International Journal of Imaging Systems and Technology. 2022; 32(6): 1846-1860.

10.

Mulmule

Kanphade

. Supervisedclassificationapproach for cervical cancer detection using Pap smear images. International Journal of Medical Engineering and Informatics (IJMEI). 2022; 14(4).

11.

Palanisamy

Athiappan

Nagalingam

. Pap smearbased cervical cancer detection using residual neural networks deep learning architecture. International Journal of Imaging Systems and Technology. 2022; 34(4): 1-16.

12.

Ontor

MZH

Ali

Hossain

Nayer

Ahmed

Bui

. YOLO_CC: Deep Learning based Approach for Early Stage Detection of Cervical Cancer from Cervix Images Using YOLOv5s Model. 2022 Second International Conference on Advances in Electrical, Computing, Communication and Sustainable Technologies (ICAECT), Bhilai, India. 2022. pp. 1-5.

13.

Bin

, et al. False Positive Reduction in Pulmonary Nodule Classification Using 3D Texture and Edge Feature in CT Images. Technology and Health Care. 2021; 29(6): 1071-1088.

14.

Shahul Mujib

Norazryana Mat

Hamidreza

. Information-based decoding of the coupling among human brain activity and movement paths. Technology and Health Care. 2021; 29(6): 1109-1118.

15.

Hanene

. A computer-aided method based on geometrical texture features for a precocious detection of fetal hydrocephalus in ultrasound images. Technology and Health Care. 2020; 28(6): 643-664.

16.

Zaid

Lahouari

. A fully-automated deep learning pipeline for cervical cancer classification. Expert Systems with Applications. 2019; 141: 1-9.

17.

Thiyaneswaran

Anguraj

Kumarganesh

Thangaraj

. Early detection of melanoma images using gray level co-occurrence matrix features and machine learning techniques for effective clinical diagnosis. International Journal of Imaging System and Technology. 2020; 1-13.

18.

William

, et al. A pap smear analysis tool (PAT) for detection of cervical cancer from pap smear images. Biomedical Engineering Online. 2019; 18(16): 1-22.

19.

Karthiga Jaya

Senthil Kumar

. Image Registration based Cervical Cancer Detection and Segmentation Using ANFIS Classifier. Asian Pacific Journal of Cancer Prevention. 2018; 9: 3203-3209.

20.

Kumarganesh

Suganthi

. An Enhanced Medical Diagnosis Sustainable System for Brain Tumor Detection and Segmentation using ANFIS Classifier. Current Medical Imaging Reviews. 2018; 14(issue.2): 271-279.

21.

Kumarganesh

Suganthi

. An efficient approach for brain image (tissue) compression based on the position of the brain tumor. International Journal of Imaging System Technology. 2016; 26(Issue 4): 237-242.

22.

Teck

Zhang

Siew

Chee

. Intelligent skin cancer detection using enhanced particle swarm optimization. Know Based Syst. 2018; 158: 118-215.

23.

Yuexiang

Linilin

. Skin lesion analysis towards melanoma detection using deep learning network. Sensors. Milan, Italy. 2018; 18(2): 1-16. doi: 10.3390/s18020556.

24.

Fei

Liu

. Fault diagnosis method of bearing utilizing GLCM and MBASA-based KELM. Sci Rep. 2022; 12: 17368.

25.

Slowik

Kwasnicka

. Evolutionary algorithms and their applications to engineering problems. Neural Comput & Applic. 2020; 32: 12363-12379.

26.

Elayaraja

, et al. An Efficient Approach for Detection and Classification of Cancer Regions in Cervical Images Using Optimization Based CNN Classification Approach. Journal of Intelligent & Fuzzy Systems. 2022; 43(1): 1023-1033. doi: 10.3233/JIFS-212871.

27.

Senthilkumar

Kumarganesh

Sivakumar

Periyarselvam

. Primitive detection of Alzheimer’s disease using neuro imaging: A progression model for Alzheimer’sdisease: Their applications, benefits, and drawbacks. Journal of Intelligent& Fuzzy Systems. 2022; 43(4): 4431-4444. doi: 10.3233/JIFS-220628.

28.

Thiyaneswaran, et al. An effective model for the iris regional characteristics and classification using deep learning alexnetwork. IET ImageProcessing. 2022; 17(1): 227-238. doi: 10.1049/ipr2.12630.

29.

Prakash

Rajiv Kannan

, et al. Meningioma brain tumor detection and classification using hybrid CNN method and RIDGELET transform. Scientific Reports. 2023; 13(14522): 1-13. doi: 10.1038/s41598-023-41576-6.

30.

Sachan

Singh

Patel

Sachan

. A study on cervical cancer screening using pap smear test and clinical correlation. Asia Pac J Oncol Nurs. 2018; 5: 337-341.

31.

Monteiro

. Design of Energy-Efficient Gaussian Filters by Combining Refactoring and Approximate Adders. 2021 IEEE International Symposium on Circuits and Systems (ISCAS), Daegu, Korea. 2021. pp. 1-5.

32.

Prasad

Gaddale

Kamath

, et al. A Study of Dimensionality Reduction in GLCM Feature-Based Classification of Machined Surface Images. Arab J Sci Eng. 2023; 1(2): 1-16.

33.

Lin

B-H

, et al. Application effect of diversified health-promoting models on rehabilitation exercises for cervical spondylotic myelopathy. Technology and Health Care. 2023; Pre-press(Pre-press): 1-14.

34.

Senthil Kumar

Kumutha

. Comparative analysis of the fuzzy C-means and neuro-fuzzy systems for detecting retinal disease. Circuits, Systems, and Signal Processing. 2020; 39(2): 698-720.

35.

Bell

Antani

Xue

Horning

Gachuhi

Wilson

Jaiswal

Befano

Long

. An observational study of deep learning and automated evaluation of cervical images for cancer screening. JNCI: Journal of the National Cancer Institute. 2019 Sep 1; 111(9): 923-32.