Computer-assisted Discrimination Of Benign And Malignant Human Tissues Based On The Dielectric Properties

The electrical properties(EPs)of biological tissues,including permittivity(?)and conductivity(?),is the the inherent properties of biological tissues under the action of the electromagnetic field.For biological tissue,the value of dielectric properties is not only related to the tissue's water content,cytosolic ion concentration,and cell membrane permeability,but also the microenvironment where the tissue is located.When there is a vicious change in the tissue,these dielectric property-related factors change,causing the dielectric properties to change.In theory,the value of dielectric properties of malignant tissues change before the change of morphology,so it may be helpful to find the suspicious cancerous tissues in the early stage.There are a large number of studies that have shown the difference between the dielectric properties of normal and malignant tissue,however,there is no clear criteria that shows the difference of the dielectric properties of them and which kind of difference can be judged as mark of cancerous tissue.Therefore,it is of great significance to study how to determine whether a tissue is cancerous according to the difference of the dielectric properties between malignant and normal tissues.There are a few literatures that begin to identify the normal and malignant tissues by using the machine learning method based on the dielectric properties.However,no literature has yet identified the benign and malignant of the human colorectal and stomach tissues based on the dielectric properties.In addition,the relevant studies have not discussed the discrimination ability of the specific parameters nor from the discrimination time to analyze.In this study,the open-ended probe method was applied to measure the dielectric properties of human colorectal and gastric tissues in the hospital operating room at the frequency range from 42.58 MHz to 500 MHz(43 frequency points in total).The measured data include 694 cases of colorectal samples and 290 cases of stomach samples.Cole-Cole model fitting was performed on the measured dielectric properties data.Receiver operating characteristics(ROC)curve analysis was used to determine the discrimination ability of each parameter.Then the parameters with high diagnostic ability were selected as the classification features,combined with support vector machine(SVM)classifier to identify the normal and malignant tissues.During the process of discrimination,we adopted three kinds of SVM parameters search methods,including improved grid search method,genetic algorithm and particle swarm optimization.We compare the results of different eigenvalue combinations and different SVM parameter optimization methods from the point of accuracy and MATLAB runtime.Considering the increasing amount of clinical data,this paper also utilized incremental SVM based on density-based algorithm.Finally,in order to make it easier to import and process data,and to visualize the implementation of the algorithm,this paper also designs a user interface.The discrimination results show that,for colorectal tissues,the discrimination accuracy of normal and malignant tissues reached more than 88%at the frequencies of 85.16 MHz and 106.45 MHz,and the SVM discrimination time based on improved grid search within 20 s.For gastric tissues,the combination of permittivity at 5 frequencies of 42.58 MHz,53.23 MHz,63.87 MHz,74.52 MHz,and 85.16 MHz achieved the highest accuracy of 84.38%,and the SVM discrimination time based on improved grid search is 3.40s.This study shows that the proposed support vector machine-aided discrimination method based on the dielectric properties of normal and malignant tissues has a potential application value.