Adaptive CNN method for prostate MR image segmentation using ensemble learning

In 2020, there were more than 1.4 million new cases of prostate cancer worldwide, and more than 375,000 deaths from the disease. The conventional diagnostic pathway hinges on the assessment of prostate specific antigen (PSA) levels and the conduct of trans-rectal ultrasound (TRUS)-guided biopsies. However, the specificity of PSA as a biomarker is notably low, at approximately 36%, due to its elevation in benign prostatic conditions, underscoring the imperative for more precise diagnostic modalities. This research leverages a dataset comprising T2-weighted magnetic resonance (MR) images from 1,151 patients, totaling 61,119 images, to refine prostate cancer diagnostics. This paper introduces methodology that utilises knowledge-based artificial intelligence (AI) frameworks with image segmentation techniques to enhance the accuracy of prostate cancer detection. The approach in this paper focuses on the segmentation of MR images into distinct anatomical zones of the prostate - specifically, the transition zone (TZ) and peripheral zone (PZ). The variations of model produce a Dice Similarity Coefficient in the range of 0.373 - 0.544 in the 95th percentile. This segmentation is critical for the automation and augmentation of diagnostic precision in prostate cancer. This approach not only aims to improve the specificity and sensitivity of prostate cancer diagnostics but also to facilitate the exploitation of publicly accessible datasets for research advancements in this domain.