Statistical atlases and machine learning tools applied to optimized prostate biopsy for cancer detection and estimation of volume and Gleason score

摘要

We discuss the use of statistical atlases and machine learning tools for determining optimized biopsy procedures. Prostate cancer diagnosis most often involves the sampling of prostate tissue via placement of a number of biopsy needles in locations that are somewhat random but try to cover the gland. The purpose of this work is to establish optimal strategies for sampling the prostate tissue, using population statistics. In particular, a statistical atlas reflecting the spatial distribution of prostate cancer has been constructed via elastic registration of expert-labeled histological 3D volumes of radical prostatectomy patients[1]. This atlas reflects the probability of encountering prostate carcinoma at a given location in the gland.