Prostate Cancer Classification Based on Histopathological Images

Abstract

Prostate cancer is a significant health concern, ranking as the third most common cancer in Malaysian men, with increasing incidence in Asia. The importance of automating the prostate cancer classification process lies in its potential to significantly improve diagnostic accuracy, reduce subjectivity, and enhance overall efficiency compared to the manual approach. The objective of this thesis is two-fold: firstly, to effectively enhance and segment crucial features in the images to aid in the classification process, and secondly, to implement a binary classification task that indicates the presence or absence of malignant tissue on histopathology images. The study compares the performance of two image enhancement approaches, stain normalization with adaptive histogram equalization (AHE) and sharpening, and stain normalization with traditional histogram equalization (HE) and sharpening. Additionally, three machine learning models, namely SVM, DenseNet121, and InceptionResNetV2, are implemented and evaluated for prostate cancer binary classification. The findings reveal that AHE contributes to better contrast enhancement and image quality preservation. Moreover, the InceptionResNetV2 model demonstrates superior performance in terms of accuracy (97.25%), sensitivity (97.5%), specificity (97.5%), and area under the curve (AUC) (97.5%).