Prostate Cancer Diagnosis Model with the Handling of Multi-Class Imbalance through the Adaptive Weighting based Deep Learning Model

Abstract

In the medical field, Prostate cancer is second leading cause of death among cancer diseases worldwide. To diagnose prostrate cancer disease, supervised machine learning and the researchers increasingly used deep learning techniques.  Despite this, machine learning and deep learning algorithms often suffer the class imbalance problem.  There is an essential need to address mult-class imbalance in diagnosing the different types of prostrate cancer diseases.  Learning from the imbalanced data is very difficult, leads to inaccurate diagnosis results.  Adaptive weighting in the deep learning model addresses the multi-class imbalance and also facilitates accurate prostate cancer disease category detection in the medical field.  The aim of the study is to develop a prostrate cancer detection model over the multi-class imbalance data using the deep learning model.  The study also proposes augmentation of structured prostrate data to enhance the deep learning-based prostrate Gleason grading classification. Conclusion: The proposed model presented the prostrate cancer diagnosis model through the risk-level classification over the imbalanced data using the deep learning model.  In the hybrid model, the data-level process can handle both the class imbalance and the data scarcity problem through data augmentation and sampling-based class sampling.  At the same time, the algorithm-level process is responsible for generating the balanced classes with the reference of improved classification accuracy based on the gradient computation.  Thus, the proposed approach effectively categorizes the multi-class prostrate cancer patients even when they are constraints in the data scarcity and class imbalance using the deep learning problem.