DOGS-2: Digital Pathology for Optimized Gleason Score-2
| Reference number | |
| Coordinator | Region Skåne - Skånes universitetssjukhus Malmö, VO Urologi |
| Funding from Vinnova | SEK 1 000 000 |
| Project duration | September 2018 - September 2020 |
| Status | Completed |
| Venture | Medtech4Health innovators |
Important results from the project
The purpose of this project was to develop a clinically applicable and reliable software for the detection and staging of prostate cancer on biopsy material. We developed and validated a sensitive AI tool for prostate biopsies, which detects cancer and assigns a Gleason grade with similar accuracy to that given by pathologists.
Expected long term effects
Our algorithm has a high sensitivity in detecting cancer areas (sensitivity: 100%, specificity: 68%). Compared with expert pathologists, the algorithm showed a high accuracy in detecting cancer areas (ICC: intra-class correlation coefficient: 0.99) and to correctly assign the Gleason grades: Gleason grade 3 and Gleason grade 4 (ICC: 0.96 and 0.94, respectively). This tool holds premises to improve reproducibility, reduce interobserver variability and speed up the diagnostic process.
Approach and implementation
The algorithm was trained on annotations performed by two experienced pathologists on 700 scanned prostate biopsy slides. The algorithm was validated on a subset of biopsy slides and showed high accuracy. Large-scale validation on a cohort of over 5000 scanned biopsy slides is currently ongoing. Once validated, the algorithm will be available to pathologists to use as an auxiliary screening tool, which will facilitate the diagnostic process.