A Follow-up Study On Prospective Para-Clinical Use by Residents of a Re-calibrating Automated Deep Learning System for Prostate Cancer Detection
Kevin Sun Zhang1, Adrian Schrader1, Nils Netzer1, Magdalena Görtz2, Viktoria Schütz2, Constantin Schwab3, Markus Hohenfellner2, Heinz-Peter Schlemmer1, and David Bonekamp1
1Department of Radiology, German Cancer Research Center, Heidelberg, Germany, 2Department of Urology, University Hospital Heidelberg, Heidelberg, Germany, 3Institute of Pathology, University of Heidelberg Medical Center, Heidelberg, Germany
Previously validated fully-automatic detection of prostate cancer by CNNs requires further prospective validation. Para-clinical case-by-case prospective prostate MRI assessment by residents was performed both before and after review of CNN probability maps superimposed on T2w images. A previously and retrospectively validated self-parametrizing nnUNet-architecture CNN trained on more than 1000 voxel-wise annotated prostate MRIs achieved ROC AUC of 0.89. Residents did not substantially change their assessment both at PI-RADS>=3 and >=4 decisions, however achieved excellent working points, indicating success of high reading capability conveyed at an expert center.
This abstract and the presentation materials are available to members only;
a login is required.