DEEP LEARNING-BASED PREDICTION OF PROSTATE CANCER USING IMAGE DATA GENERATOR CLASSES

Abstract

Prostate cancer is a common cancer among men, with the highest rate being recorded among men above 40 years of age. The correct diagnosis is a critical, time-consuming process that histopathologists have traditionally conducted through systematic investigation of biopsy samples. Although this is necessary for reliable detection, it is usually tedious and prone to interobserver error. The use of the latest medical imaging and computational technologies has significantly improved histopathologists' ability to detect and grade prostate cancer. Diagnosis of prostate cancer may be performed through several clinical and non-clinical imaging tests, e.g., multiparameter magnetic resonance imaging (mpMRI). Nevertheless, specific diagnostic methods continue to have high false-positive and false-negative rates, thereby limiting their accuracy. To overcome these issues, this paper uses an image data augmentation approach based on the Image Data Generator class to increase the diversity of the training data and enhance model generalization. Deep learning architectures, such as VGG-16, ResNet-50, and DenseNet121, are used to analyze prostate biopsy images and assess their performance in detecting cancer. The selection of these architectures is based on their demonstrated performance across a variety of benchmark datasets, and they are being applied to biopsy images to enhance the accuracy and resilience of their diagnostic results.