Predicting Biochemical Recurrence from Prostatectomy Slides - the LEOPARD Challenge

Khrystyna Faryna, Clément Grisi, Nefise Uysal, Hans Pinckaers, Joep Bogaerts, Solene-Florence Kammerer-Jacquet, Pierre Allaume, Vittorio Agosti, Paulo Guilherme de Oliveira Salles, Yuri Tolkach, Reinhard Buettner, Sofiya Semko, Maksym Pikul, Axel Heidenreich, Adam Kowalewski, Leslie Tessier, Frédérique Meeuwsen, Jolique van Ipenburg, Till Nicke, Johannes Lotz et al. (31 additional authors not shown)

Keywords: Prostate Cancer, Biochemical Recurrence, Deep Learning, AI, Histopathology

TL;DR: LEOPARD is a benchmark study and a challenge evaluating AI-based algorithms predicting biochemica recurrence from prostatectomy slides.

Registration Requirement: Yes

Abstract: Prostate cancer affects over 1.4 million men yearly, with approximately 30% experiencing biochemical recurrence (BCR) after prostatectomy. Current clinical tools predict risk at the population level but are less accurate for individuals. The LEarning biOchemical Prostate cAncer Recurrence from histopathology sliDes (LEOPARD) challenge benchmarked AI models to predict time to BCR directly from H&E-stained prostatectomy slides using 2,181 cases from four countries. Sixteen teams participated; nine models were selected for final evaluation. Top AI models achieved a C-index of 0.740, comparable to histopathological grading (0.739) and slightly below Cancer of the Prostate Risk Assessment Post-surgical (CAPRA-S) - 0.785. Combining AI with histopathological grading improved performance to 0.766, and with CAPRA-S to 0.799, significantly enhancing BCR risk prediction.

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 108