Computer-Aided Diagnostics of Prostate Cancer with Automated Image Analysis using Machine Learning

AttachmentSize
Full Paper924.56 KB

Abstract
Prostate cancer is the most common cancer among men, other than skin cancer. Prostate cancer screening is done with a blood test that measures levels of prostate-specific antigens (PSA) in the blood and by palpating the prostate through the rectum. If indications are found in either test, the next step is a biopsy of the prostate, which involves using one or more thin needles to remove cores from multiple areas of the prostate. Three sections are taken from each core, and each section is prepared on a glass slide. A pathologist then renders a final diagnosis (i.e. benign vs cancer) after examining each of the slides though a microscope. Because the number of samples can be high per patient, and a given pathologist may see many patients’ biopsies in a day, prostate biopsies are one of the most time consuming case types pathologists must process. Here we show how an automatic system can assist the pathologist in prostate cancer diagnosis by sifting through numerous samples and flagging those suspicious for cancer. Experimental results have produced repeatable accuracies averaging as high as 85%.