Abstract #4376

Does Machine Learning, As An Independent Arbitrator Of MR Contrast-Ranking In Prostate Cancer Exams, Agree With PI-RADS version 2?

Steve Patterson¹, Peter Lee², Chris V. Bowen^3,4,5, Jennifer Merrimen⁶, Cheng Wang⁶, Steven D. Beyea^3,4,5, and Sharon E. Clarke^3,4,5

¹Steve Patterson, Nova Scotia Health Research Foundation, Halifax, NS, Canada, ²Faculty of Computer Science, Dalhousie University, Halifax, NS, Canada, ³Diagnostic Radiology, Dalhousie University, Halifax, NS, Canada, ⁴Biomedical Translational Imaging Centre, Nova Scotia Health Authority, Halifax, NS, Canada, ⁵Physics and Atmospheric Science, Dalhousie University, Halifax, NS, Canada, ⁶Anatomical Pathology, Dalhousie University, Halifax, NS, Canada

We show that a simple machine learning algorithm validated most, but not all, aspects of the Prostate Imaging Reporting and Data System (PI-RADS) version 2 formalism derived exclusively from clinical perspectives. Specifically, the value of diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) sequences in the peripheral zone was confirmed. In contradistinction to PI-RADS, DWI was found to be more valuable in the transition zone than T2 weighted imaging; however, a T2 texture feature afforded a small but significant increase in classifier accuracy in this zone.

This abstract and the presentation materials are available to members only; a login is required.

Join Here