Comparison of Prostate Cell Image Classification Using CNN: ResNet-101 and VGG-19

Abstract

Being the most common disease in men, prostate cancer attacks the urinary system. Men with higher androgen levels have a greater risk of developing prostate cancer. This cancer occurs in the prostate gland of the male reproductive tract. This cancer appears when it begins to mutate and reproduce uncontrollably. Risk factors for prostate cancer include age, race and family history. This study classified prostate cell images based on their severity. Along with today’s technological advancement, especially research on image classification, it will be simpler for medical personnel to educate the public on how to recognize the severity of prostate cancer through a system. This image classification system utilized the preserved models of the deep learning method from the Convolutional Neural Network (CNN) algorithms: ResNet-101 and VGG-19. It aims to discover the most appropriate algorithm in image classification, determined from the training graph and confusion matrix calculation results. The measurements of the models’ reliability encompassed accuracy, computation time, graph stability, and confusion matrix calculation results. ResNet-101 appeared as the model with the highest training data accuracy and fastest computation time. The confusion matrix calculation also unveiled that ResNet-101 acquired the greatest results with an average accuracy of 97.70%, precision of 93.19%, recall of 93.25%, specificity of 98.62% and F-score of 93.11%, demonstrating its superiority over VGG-19 in classifying prostate cell images based on testing data.