Article Dans Une Revue Discover Oncology Année : 2025

Prediction of biochemical prostate cancer recurrence from any Gleason score using robust tissue structure and clinically available information

Résumé

Biopsy information and protein Prostate-Specific Antigen (PSA) levels are the most robust information available to oncologists worldwide to diagnose and decide therapies for prostate cancer patients. However, prostate cancer presents a high risk of recurrence, and the technologies used to evaluate it demand more complex resources. This paper aims to predict Biochemical Recurrence (BCR) based on Whole Slide Images (WSI) of biopsies, Gleason scores, and PSA levels. A U-net model was used to segment phenotypic features and trained on images from the Prostate Cancer Grade Assessment (PANDA) database to segment tumorous regions from pre-processed and scored WSI of biopsies. Then, the model was tested on data from publicly available repositories achieving an Intersection over Union of 87%. Tissue features, Gleason scores, and PSA levels provided high accuracy and precision in classifying patients according to their risk of presenting recurrence, for any Gleason score sampled. The trained classifier model demonstrated a 79.2% relative accuracy, and a precision of 69.7% for patients experiencing recurrences before 24 months. Our results provide a robust, cost-efficient approach using already available information to predict the risk of BCR.
Fichier principal
Vignette du fichier
s12672-025-01896-7.pdf (2) Télécharger le fichier
Origine Publication financée par une institution

Dates et versions

hal-04942670 , version 1 (12-02-2025)

Identifiants

Citer

Laura E Marin, Daniel I Zavaleta-Guzman, Jessyca I Gutierrez-Garcia, Daniel Racoceanu, Fanny L Casado. Prediction of biochemical prostate cancer recurrence from any Gleason score using robust tissue structure and clinically available information. Discover Oncology, 2025, 16 (1), pp.128. ⟨10.1007/s12672-025-01896-7⟩. ⟨hal-04942670⟩
0 Consultations
0 Téléchargements

Altmetric

Partager

More