Investigating Cervical Cancer Detection Frameworks Based on Machine Learning: The Critical Tradeoff Between Accuracy and Data Security ^†

Abstract

Cervical cancer has emerged as the most prevalent and deadly illness affecting women across the globe. Researchers are trying their best to detect this life-threatening illness accurately. In view of this only, machine learning approaches, multiple medical procedures, statistical models, etc., have been utilized to provide optimized and efficient treatment to all patients to protect their lives. In this study, we have compared previously proposed frameworks for the early detection of cervical cancer and analysis of patients’ data security. We demonstrated the respective benefits and limitations, investigated the datasets and the type of data employed, and analyzed the accuracy of the healthcare procedures utilized for patients in terms of improving management. The limitations of reviewed studies show that more reliable proposals need to be presented by researchers in future. Based on this only, it is concluded that the accurate and early detection of cervical cancer shows a tradeoff with patients’ data security while communicating across healthcare institutions.

1. Introduction

Nowadays, cervical cancer is generally found to be a very commonly occurring life-threatening disease. It completely threatens females’ physical as well as mental health. It starts with the growth of cells in the cervix area of women, connected to the female vagina [1]. In the view of this particular life-threatening among the females, researchers are working to ensure the early detection as well as prevention of the disease among women. Regular examination and early detection are required to save women’s lives as they face multiple symptoms at first. These initial symptoms further continue to spread to other body organs like the lungs, lymph nodes, bones and liver. Symptoms like blockage in tubes carrying urine, vaginal bleeding, pelvic pain, weight loss and leg pain are the major indicators of cervical cancer [2]. These symptoms can be monitored by healthcare practitioners by utilizing their concerned practices. Moreover, unavoidable discomfort after puberty and vaginal hemorrhage after coitus, between periods of menstruation, and after gynecological examination are considered major indications of cervical cancer.

Further, digital and enhanced technologies or procedures can be utilized to ensure prompt diagnosis before the disease can reach this cancerous condition. In this respect only, technologies like machine learning, artificial intelligence, deep learning, blockchain, etc., are utilized in the healthcare sector more frequently. Machine learning is defined as a subset of AI (artificial intelligence) which deals with the three different problems, i.e., clustering, regression, and classification, in a significant way [3]. This technology utilizes algorithms as well as data in order to train the users as well as sector practitioners for the major improvement of accuracy. The algorithms consume data to recognize as well as learn involved patterns so that this information can be used further for classification task [4]. Machine learning works on supervised learning as well as unsupervised learning. The early prediction of cervical cancer can be performed with the supervised machine learning approach.

Further, supervised learning refers to the process of mapping input variables with an output variable and applying this mapping in order to predict the results from the entire dataset [5], as shown in Figure 1. One of the best examples of supervised learning is SVM (support vector machine). It is a very powerful algorithm that is used for nonlinear classification, linear classification, the detection of outliers present in the data, and regression [6]. It can be helpful in many applications like anomaly detection, face detection, gene expression analysis, handwriting identification, spam detection, image classification, and text classification. It focuses on identifying the separating hyperplane among the present classes in the targeted feature. In the same manner, other machine learning algorithms can be particularly helpful in generalizing the actual condition patient in any disease such as cervical cancer.

Figure 1. Supervised learning approach in machine learning.

In the healthcare sector, patients’ data is considered as most sensitive data while transmitting it from one department to another. In this case, blockchain is one of the enhanced technologies used to provide better security to all patients as well as healthcare practitioners. Blockchain helps to transfer the digital assets in the form of bitcoin [7]. This bitcoin transaction is broadcasted into each and every peer member connected into the network. Clients communicate by utilizing cryptography algorithms. Transactions are represented in the block and then this block is further broadcast to all nodes present in the blockchain network. After the approval of the transaction, all information is further encrypted by using hashing techniques as shown in Figure 2. In this manner, blockchain helps in providing enhanced security to each and every sector.

Figure 2. Process of blockchain.

2. Objectives of the Study

This study has been conducted on the basis of the objectives discussed below:

• Reviewing the previously proposed framework for the detection of cervical cancer in terms of accuracy as well as data security while communicating across healthcare institutions;
• Comparing the reviewed framework with the help of techniques, algorithms, tools, a utilized dataset, and data types, and in terms of benefits, limitations, accuracy, and data security.
• Analyzing the results and research gaps that must be considered to provide foolproof frameworks for the early, secure, and accurate diagnosis of cervical cancer.

3. Literature Review

This section discusses the early prediction or detection of cervical cancer among women and the security of sensitive healthcare data. Based on this, many researchers have already conducted research and provided desired results, as per details discussed below.

Researchers proposed a framework by utilizing federated machine learning to predict cervical cancer among females. It included processes like data redundancy, data cleaning, and data augmentation. This proposed framework identified items with 99.26% of the accuracy of other traditional results [8]. The major benefit of the technique is the adoption of an additional security layer in blockchain technology. This study has a significant limitation, which relates to the use of the proposed framework on a larger scale.

Cervical cancer is commonly found to be a life-threatening disease, causing premature deaths among females. In view of this alone, researchers proposed a novel framework a utilizing hybrid feature selection approach in order to predict the cervical cancer. This predictive framework, along with recent risk trends and early screening, investigates the actual situation of patients suffering with cervical cancer. It has been found that the framework is able to achieve 99.19% accuracy overall along with 100% sensitivity by integrating PCA (Principal Component Analysis), XGBoost classifier, RF (Random Forest), and MLP (Multilayer Perceptron) [9].

Recently, it was also found that researchers proposed a novel framework named as CACCD-GOADL (computer-aided cervical cancer diagnosis utilizing gazelle optimizer algorithm with deep learning). The main objective of conducting this study is to detect cervical cancer by using the images as a dataset [10]. It utilized an SELM (stacked extreme learning machine) methodology along with the improved MobileNetv3 systems. Further, it is found that this method is able to achieve best accuracy of 99.38%, precision of 96.71%, recall of 97.45% and F1-score of 97.04 [10]. The limitation of this respective research is that it does not discuss anything related to the security and privacy of the sensitive data of patients and other healthcare practitioners.

Cervical cancer has emerged as a serious concern among females. Delays in detection as well as treatment can cause a devastating impact on the health of patients. Avoiding time-consuming techniques and other conventional procedures has always resulted in the efficient and early detection of this life-threatening condition. Researchers have proposed a framework for the detection of cervical cancer using machine learning in the integration of blockchain technology from any kind of authorized access of users to healthcare data. It is found that study can achieve only 93% accuracy [11].

Researchers have also developed a framework to analyze the cancer cells using machine learning approaches. Data was collected by using the Internet of Things and further communicated with the help of blockchain technology. It has been found that the method is able to achieve nearly 90.63% accuracy by utilizing linear discriminant analysis (LDA) and support vector machine (SVM) classifier approaches [12]. In this study, blockchain technology is helpful in training the data for the improved protection of patients from any kind of intruder disturbance throughout the process.

Further, a novel framework incorporates two different data-balancing approaches, i.e., Adaptive Synthetic Sampling and Synthetic Minority Oversampling techniques, in order to discard issues like data imbalance. Discussing machine learning algorithms, study has utilized support vector machine, K-nearest neighbors, random forest, naïve Bayes, logistic regression, decision tree, etc. It is found that researchers can achieve nearly 97% accuracy and precision [13]. The major limitation is seen regarding the computational cost and utilized dataset.

In order to discard all the issues related to the detection of cancers and medical diseases, researchers developed a framework named digital twin. It is an automated application for the early detection of cervical cancer. The proposed framework has achieved impressive accuracy of 98.91% by utilizing large dataset with 1013 images [14]. The major advantage of using this newly developed framework is seen with the reduction in medical expense, an extension in actual life expectancy, an improvement in health, etc. However, this particular research has not discussed anything related to blockchain technology and security of data while communicating with the multiple healthcare institutions [14].

A novel framework, Web Framework for Cervical Cancer Detection System (WFC2DS), is used to update the diagnosis system of cervical cancer. It integrates multiple machine learning algorithms, i.e., decision tree, support vector machine, random forest classifier, k-nearest neighbor, adaboost, and artificial neural network. The maximum accuracy achieved in the early diagnosis of cervical cancer was 98.1% [15]. There are no such data-balancing techniques are utilized by the researchers for the diagnosis of medical diseases.

Rather than just considering the detection of cancer, data security is one of the major concerns of healthcare institutions. A new framework is proposed to ensure the secure detection of cancer by using Internet of Medical Things, blockchain systems, and machine learning approaches. It utilized GRUs (Gated Recurrent Units) along with RNN (Recurrent Neural Networks) for data collection and achieving the desired results. The actual accuracy of the developed framework is 95% [16]. The main limitation is that researchers did not utilize the real-time datasets for the desired results.

Discussing other types of cancers, it is seen that researchers can achieve good percentiles of accuracy in terms of the actual diagnosis of cancer among patients, along with ensuring data security while communicating across multiple healthcare institutions. Researchers are able to achieve accuracy values like 97% for breast cancer [17], 98.8% for pancreatic cancer [18], 98.9% for breast cancer [19], and 99.92% for the diagnosis of cancerous brain tumors [20]. Each framework has displayed respective limitations as well as benefits. However, it is seen that all of the frameworks have addressed security concerns. In [20], the authors consider enhanced embedded encryption approaches for secure communication and the transmission of data across the multiple healthcare institutions. No doubt, this can also be improved further with the addition of AI-assisted frameworks for digital security of sensitive data associated with the patients as well as all other healthcare practitioners. Transmission of patients’ data can result into cyberattacks over the communicating channels. In this context, researchers have proposed a survey to analyze botnet attacks [21]. Some of the researchers have also proposed the framework by utilizing Internet of Medical Things technology in order to improve the data security as well as efficient data aggregation. Further, it has resulted into improved computational cost, storage, resilience, energy consumption, communication cost etc. [22]. In the further researches, another type of clinical issues like pressure ulcers are discussed by using deep learning algorithms [23].

4. Comparison Table

This section shows the comparison among reviewed frameworks for the detection of cervical cancer on the basis of their utilized technique, algorithm, tool, dataset, data type, benefit, limitation, accuracy, and data security. Accuracy and data security shows the actual performance of frameworks in terms of diagnosis as well as security of patients, as shown in Table 1.

Table 1. Comparative analysis of review frameworks based on different parameters.

Techniques	Algorithm	Tool	Data Set	Data Type	Benefit	Limitation	Accuracy	Data Security	Ref.
Federated machine learning in the integration of Blockchain and Internet of Medical Things	Algorithms based on Scaled conjugate gradient, Levenberg Marquardt, and Bayesian regularization	MATLAB	UCI cervical cancer risk factors	Numeric	Study includes the security as well as privacy of healthcare sensitive data by getting feedbacks from the respective individuals	Proposed study required to be utilized at large scale.	99.26%	Yes	[8]
Random oversampling and ensemble Machine Learning method i.e., hard voting	Algorithms based on machine learning classifiers	Jupyter Notebook and Google Colab	UCI Machine Learning repository	Numeric	Able to found best results with the combination of PCA, XGBoost, RF and MLP in the terms of accuracy and sensitivity	Unable to provide best results as an online screening tool along with security	99.19%	No	[9]
CACCD-GOADL (Computer Aided Cervical Cancer Diagnosis utilizing Gazelle Optimizer Algorithm with deep learning)	SELM methodology along with MobileNetv3 system	Not defined	Herlev dataset	Images	Better results have found in terms of accuracy, precision, recall and F1-score as compared to other approaches.	No discussion regarding the security as well as privacy of the data	99.38%	No	[10]
Machine learning and blockchain smart contracts	Automated prediction and secure algorithms	Graphical tool	Repositories	Numeric	Classification approaches are used to remove issues like over-fitting, outliers, bias present in the dataset and to provide more promising result.	Framework accuracy decreases with nearly by 6% after training of dataset	93%	Yes	[11]
Support Vectors and structure feature selection	Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM)	Not defined	Primary source dataset gathered from Kilpauk Medical College and Hospital, Government of Tamil Nadu	Images	Best classification results have found by utilizing machine learning approaches, Internet of Things and blockchain technology	Blockchain technology has not integrated with the deep learning algorithms for the secure communication over IoT network	90.63%	Yes	[12]
Adaptive Synthetic Sampling and Synthetic Minority Oversampling	Support vector machine, k-nearest neighbor, random forest, naïve bayes, logistic regression, decision tree	Not defined	UCI machine learning repository	Numeric	Early detection of cervical cancer utilizing explainable Artificial Intelligence and feature selection	High computational cost, no data security and utilization of smaller datasets.	97%	No	[13]
Digital Twin application	Cervix classifier model	Keras, Tensorflow, Sklearn package, python were utilized for the development along with Numpy and Pandas libraries.	Multi-cell dataset i.e., SIPaKMeD	Images	Reduction in the medical expenses, extension in the actual life expectancy, improvement in the health	No data security has considered by researchers in the respective proposal	98.31%	No	[14]
Web framework for cervical cancer detection system	Machine learning algorithms	Pandas dataframe	UCI machine learning repository	Numeric with 36 attributes	Proposed framework is not just bound with providing results in diagnosis of cervical cancer	No data security has considered by researchers in the respective proposal	98.1%	No	[15]
Internet of Medical Things, machine learning and blockchain	Gated recurrent units with neural networks	Not defined	Kaggle	Numeric	Detection of cancer is possible along with the data security by using blockchain systems and highly enhanced AES cryptosystems.	Standard datasets are utilized	95%	Yes	[16]
Privacy preserving detection framework	Deep learning approaches	Python	kaggle	Images13]	Utilized enhanced integrated encryption algorithms for the secure detection of medical condition	Additional AI-assisted security frameworks must be involved.	99.92%	Yes	[20]

5. Discussion and Gap Analysis

This section presents the graphical analysis of the previously proposed frameworks. From Table 1, it can be concluded that researchers are obtaining the desired results for the detection of cervical cancer in terms of accuracy. It is found that the numeric type of dataset provides the better results as compared to image datasets. No doubt, Ref. [8] can to provide the reliable detection of cervical cancer among women, with an accuracy of 99.26%, along with the integration of data security mechanisms, as shown in Figure 3. Refs. [9,11,13,15,16] utilized a numeric type of data; they showed a detection accuracy of less than 99.26%, as shown in Figure 3. Moreover, a tradeoff is seen between data security and the detection of cervical cancer among the women. After analyzing the previously proposed cervical cancer frameworks, some of the frameworks based on the detection of other types of cancer are analyzed. It is found that research regarding other cancer type, i.e., breast cancer is not facing any tradeoff between the detection accuracy and data security. In [20], it is seen that 99.92% of the accuracy was achieved by the researchers along with the integration of data security mechanisms for secure communication across healthcare institutions. So, in the view of cervical cancer, the major research gap is regarding the tradeoff between data security and detection accuracy.

Figure 3. Accuracy analysis along with data security for detection of cervical cancer [8,9,10,11,12,13,14,15,16,20].

6. Conclusions and Future Scope

Cervical cancer has emerged as a life-threatening medical disease. Researchers have kept on proposing novel frameworks for its early detection among women. In this study, it is concluded that researchers are providing better and more desirable results on the basis of accurate detection of cervical cancer. However, discussing data security, we showed a tradeoff between accurate detection and data security while communicating across healthcare institutions. Further, to achieve both accurate diagnosis and data security simultaneously for cervical cancer, researchers need to use enhanced mechanisms in the future.