Pathologist-like explainable AI for interpretable Gleason grading in prostate cancer

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: <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub><mml:mrow><mml:mn>0.713</mml:mn></mml:mrow> <mml:mrow><mml:mo>±</mml:mo> <mml:mn>0.003</mml:mn></mml:mrow> </mml:msub> </mml:math> vs. <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub><mml:mrow><mml:mn>0.691</mml:mn></mml:mrow> <mml:mrow><mml:mo>±</mml:mo> <mml:mn>0.010</mml:mn></mml:mrow> </mml:msub> </mml:math> ) 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.

Authors

Institutions

Publication Details

Journal
Nature Communications
Published
2025-10-08
DOI
https://doi.org/10.1038/s41467-025-64712-4
Citations
5
Primary Topic
AI in cancer detection
Type
article
Field-Weighted Citation Impact
12.40

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Pathologist-like explainable AI for interpretable Gleason grading in prostate cancer

Sean R. Williamson, Andreas Gocht, Sara Laiouar-Pedari, Jula Westhoff et al.
5 citations
Nature Communications
AI in cancer detection
12.40
article

Pathologist-like explainable AI for interpretable Gleason grading in prostate cancer

Sean R. Williamson, Andreas Gocht, Sara Laiouar-Pedari, Jula Westhoff, Nidia Messias, Maximilian Seidl, Tabea-Clara Bucher, Guoping Cai, Oudai Hassan, Kiril Trpkov, Yu Chun Tam, Felix Bremmer, Gilbert Georg Klamminger, Christian M. Schürch, Jaroslaw Augustyniak, Hendrik A. Mehrtens, Jon Griffin, Niels J. Rupp, Francesca Sanguedolce, Maria Sarah L. Lenon, Constanze Döring, Carolin Victoria Schneider, Varsha Manucha, Nicole Swarbrick, Ali Amin, Tim Holland‐Letz, Monika Haemmerle, Gabriele Deubler, Tirtha Chanda, Kyung Park, Kirsten D. Mertz, Michael Brodhun, Irene Krücken, Florian Weber, Thomas Kalinski, Rainer Grobholz, Yuri Tolkach, Alessia Cimadamore, Éva Compérat, Marcus Bauer, Christian Hörner, Rajal B. Shah, Angelique Levi, Fei Chen, Farshid Siadat, Christoph Wies, Tri Q. Nguyen, Andreas Erbersdobler, Wiebke Solaß, Jakob Nikolas Kather, Hiroshi Miyamoto, Nadine Therese Gaisa, Sarah Haggenmüller, Ezra Baraban, Christina Neppl, Nazeel Ahmad, Titus Josef Brinker, Tilman T. Rau, Ahmed Shabaik, Liang Cheng C, Michelle R. Downes, Gesa Mittmann, Liwei Jia, Ulrich Sommer, Mathias Wagner, Diana Taheri, Siamak Sakhaie
article en
5 citations

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: <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub><mml:mrow><mml:mn>0.713</mml:mn></mml:mrow> <mml:mrow><mml:mo>±</mml:mo> <mml:mn>0.003</mml:mn></mml:mrow> </mml:msub> </mml:math> vs. <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub><mml:mrow><mml:mn>0.691</mml:mn></mml:mrow> <mml:mrow><mml:mo>±</mml:mo> <mml:mn>0.010</mml:mn></mml:mrow> </mml:msub> </mml:math> ) 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.

Nature CommunicationsVol. 16(1)
University of Foggia (IT), Isfahan University of Medical Sciences (IR), University of Bern (CH), University of Santo Tomas Hospital (PH), University of Udine (IT), University of Leeds (GB), Cleveland Clinic (US), Sunnybrook Health Science Centre (CA), Johns Hopkins University (US), Henry Ford Health System (US), University of Mannheim (DE), German Cancer Research Center (DE), University of Cologne (DE), Universität Ulm (DE), University of Zurich (CH), Washington University in St. Louis (US), Heidelberg University (DE), Brown University (US), Frankston Hospital (AU), Rockyview General Hospital (CA), University Hospital Heidelberg (DE), Health Sciences Centre (CA), University Hospital of Basel (CH), Jackson Memorial Hospital (US), University of Rochester Medical Center (US), University of Mississippi Medical Center (US), University Medical Center Utrecht (NL), Yale University (US), University of California San Diego (US), NYU Langone Health (US), James A. Haley Veterans' Hospital (US), Düsseldorf University Hospital (DE), Städtisches Klinikum Karlsruhe (DE), Schwyz University of Teacher Education (CH), University Hospital of Zurich (CH), National Center for Tumor Diseases (DE), University Hospital Schleswig-Holstein (DE), Universitätsklinikum Aachen (DE), Technische Hochschule Georg Agricola (DE), Kreiskliniken Reutlingen (DE), Department of Biomedicine Basel (CH), Helios Klinikum Erfurt (DE), Westsächsische Hochschule Zwickau (DE), Klinikum Bremen-Mitte (DE), University Hospital Cologne (DE), University Hospital Ulm (DE), University Medical Centre Mannheim (DE), Universitätsklinikum Brandenburg an der Havel (DE), Kantonsspital Aarau (CH), University Hospital Carl Gustav Carus (DE), Pathwest Laboratory Medicine (AU), National Kidney and Transplant Institute (PH), Heinrich Heine University Düsseldorf (DE), University of Rostock (DE), University of Regensburg (DE), Martin Luther University Halle-Wittenberg (DE), Tehran University of Medical Sciences (IR), Medical University of Vienna (AT), University of Tübingen (DE), The University of Texas Southwestern Medical Center (US), RWTH Aachen University (DE), Ruhr University Bochum (DE), University of Sheffield (GB), Saarland University (DE), University of Lübeck (DE)
Pfizer, AstraZeneca, National Institute for Health and Care Research, Department of Health and Social Care, European Commission, Deutscher Akademischer Austauschdienst, Deutsche Forschungsgemeinschaft, Eisai, Deutsche Krebshilfe, Gemeinsame Bundesausschuss, National Institutes of Health, HORIZON EUROPE Framework Programme, Leeds Biomedical Research Centre, Daiichi Sankyo Europe
Openalex Percentile: Top 2%
AI in cancer detection
12.40
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.