Artificial Intelligence for Medical Image Interpretation Using Expert Knowledge and Machine Learning

Abstract

In 2022 268,490 new cases and 34,500 deaths was estimated for prostate cancer in the United States. Diagnosis of prostate cancer is primarily based on prostate-specific antigen (PSA) screening and trans-rectal ultrasound (TRUS)-guided prostate biopsy. PSA has a low specificity of 36% since benign conditions can elevate the PSA levels. The data set used for prostate cancer consists of t2-weighted MR images for 1,151 patients and 61,119 images. This paper presents an approach to applying knowledge-based artificial intelligence together with image segmentation to improve the diagnosis of prostate cancer using publicly available data. Complete and reliable segmentation into the transition zone and peripheral zone is required in order to automate and enhance the process of prostate cancer diagnosis.