Multi-class classification of prostate cancer MRIS based on UCLA score using deep transfer learning models

Abstract

Prostate cancer is the most frequent cancer in the men population, and early detection is critical in order to lower mortality rates from the disease. With its superior soft-tissue contrast, magnetic resonance imaging (MRI) has become the imaging method of choice for the localization of PCa. In terms of diagnosing PCa of the transition zone, T2-weighted images are the most useful tool among the several MRI modalities. In this proposed model, the PCa is classified based on the T2w MRI data. The proposed model is a deep learning approach, which includes the deep transfer learning models for the classification of PCa. For classifying the data, the different variants of VGG-16, VGG-19, and MobileNet-v3Large transfer learning models are used. These models are modified using different optimizers for varying the learning rate. Optimizers like Adam, AdaMax, SGD, RMSprop, and Ftrl are used in this research. For evaluation, the ampMRI dataset with 845 patient records with unique "UCLA" scores of the ROI was used for multi-class classification. For performance analysis, accuracy, sensitivity, specificity, precision, and F1 score are computed based on the classification. Finally, according to the results, the performance was compared among the different proposed models for validation. The proposed models optimized using the Ftrl optimizer have obtained better performances with 93.31% accuracy, 93.92% accuracy, and 95.27% accuracy for VGG-16-Model-04, VGG-19-Model-04, and MobileNet-v3 respectively.