Mittmann, Gesa and Laiouar-Pedari, Sara and Mehrtens, Hendrik A. and Haggenmuller, Sarah and Bucher, Tabea-Clara and Chanda, Tirtha and Gaisa, Nadine T. and Wagner, Mathias and Klamminger, Gilbert Georg and Rau, Tilman T. and Neppl, Christina and Comperat, Eva Maria and Gocht, Andreas and Haemmerle, Monika and Rupp, Niels J. and Westhoff, Jula and Krucken, Irene and Seidl, Maximilian and Schurch, Christian M. and Bauer, Marcus and Solass, Wiebke and Tam, Yu Chun and Weber, Florian and Grobholz, Rainer and Augustyniak, Jaroslaw and Kalinski, Thomas and Horner, Christian and Mertz, Kirsten D. and Doring, Constanze and Erbersdobler, Andreas and Deubler, Gabriele and Bremmer, Felix and Sommer, Ulrich and Brodhun, Michael and Griffin, Jon and Lenon, Maria Sarah L. and Trpkov, Kiril and Cheng, Liang and Chen, Fei and Levi, Angelique and Cai, Guoping and Nguyen, Tri Q. and Amin, Ali and Cimadamore, Alessia and Shabaik, Ahmed and Manucha, Varsha and Ahmad, Nazeel and Messias, Nidia and Sanguedolce, Francesca and Taheri, Diana and Baraban, Ezra and Jia, Liwei and Shah, Rajal B. and Siadat, Farshid and Swarbrick, Nicole and Park, Kyung and Hassan, Oudai and Sakhaie, Siamak and Downes, Michelle R. and Miyamoto, Hiroshi and Williamson, Sean R. and Holland-Letz, Tim and Wies, Christoph and Schneider, Carolin V. and Kather, Jakob Nikolas and Tolkach, Yuri and Brinker, Titus J. (2025) Pathologist-like explainable AI for interpretable Gleason grading in prostate cancer. NATURE COMMUNICATIONS, 16 (1): 8959. ISSN , 2041-1723
Full text not available from this repository. (Request a copy)Abstract
The aggressiveness of prostate cancer is primarily assessed from histopathological data using the Gleason scoring system. Conventional artificial intelligence (AI) approaches can predict Gleason scores, but often lack explainability, which may limit clinical acceptance. Here, we present an alternative, inherently explainable AI that circumvents the need for post-hoc explainability methods. The model was trained on 1,015 tissue microarray core images, annotated with detailed pattern descriptions by 54 international pathologists following standardized guidelines. It uses pathologist-defined terminology and was trained using soft labels to capture data uncertainty. This approach enables robust Gleason pattern segmentation despite high interobserver variability. The model achieved comparable or superior performance to direct Gleason pattern segmentation (Dice score: 0.713 +/- 0.003\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${0.713}_{\pm 0.003}$$\end{document} vs. 0.691 +/- 0.010\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${0.691}_{\pm 0.010}$$\end{document}) while providing interpretable outputs. We release this dataset to encourage further research on segmentation in medical tasks with high subjectivity and to deepen insights into pathologists' reasoning.
| Item Type: | Article |
|---|---|
| Uncontrolled Keywords: | ISUP CONSENSUS CONFERENCE; INTEROBSERVER-REPRODUCIBILITY; ARTIFICIAL-INTELLIGENCE; INTERNATIONAL SOCIETY; DIAGNOSIS; ADENOCARCINOMA; |
| Subjects: | 000 Computer science, information & general works > 004 Computer science 600 Technology > 610 Medical sciences Medicine |
| Divisions: | Medicine > Lehrstuhl für Pathologie |
| Depositing User: | Dr. Gernot Deinzer |
| Date Deposited: | 23 Mar 2026 11:37 |
| Last Modified: | 23 Mar 2026 11:37 |
| URI: | https://pred.uni-regensburg.de/id/eprint/68105 |
Actions (login required)
 |
View Item |
