Meeting Banner
Abstract #1828

A U-Net applied to diffusion-weighted images compared to full multiparametric clinical PI-RADS assessment for detection and segmentation of significant prostate cancer

Patrick Schelb1, Simon Kohl2, Jan-Philipp Radtke1,3, Markus Hohenfellner3, Heinz-Peter Schlemmer1, Klaus Maier-Hein2, and David Bonekamp1

1Radiology, German Cancer Research Center, Heidelberg, Germany, 2Medical Informatics, German Cancer Research Center, Heidelberg, Germany, 3Urology, University Hospital Heidelberg, Heidelberg, Germany

A U-Net applied to diffusion-weighted imaging (DWI) only was trained with 3T MRI data from a single system in 316 consecutive patients. All clinical MR lesions were targeted with fusion biopsy in addition to extended 24-core systematic biopsy. The performance of the final CNN ensemble on the test set achieved comparable sensitivity in comparison to multiparametric clinical assessment and demonstrated the method’s ability to generate stable results in an unseen subset. These findings highlight the ability of computer vision to closely model the clinical task with fewer data and encourage development of the method in larger cohorts.

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

Join Here