Related Experiment Video
Updated: Jul 19, 2025
08:08
Detection and Isolation of Cancer in Prostate Biopsies Using Stimulated Raman Histology and Artificial Intelligence
Published on: June 10, 2025
92
An international multi-institutional validation study of the algorithm for prostate cancer detection and Gleason
Yuri Tolkach1, Vlado Ovtcharov2, Alexey Pryalukhin3
1Institute of Pathology, University Hospital Cologne, Cologne, Germany. yuri.tolkach@gmail.com.
NPJ Precision Oncology
|August 15, 2023
Summary
A new AI tool accurately detects prostate cancer (PCA) in biopsies and assigns Gleason grades, matching expert pathologist performance. This technology streamlines pathology workflows, improving diagnostic efficiency and accuracy for prostate cancer detection.
Area of Science:
- Computational pathology
- Artificial intelligence in diagnostics
- Urologic pathology
Background:
- Prostate biopsy analysis is labor-intensive, involving numerous slides per case.
- Accurate prostate cancer (PCA) detection and Gleason grading are critical for patient management.
- Existing diagnostic methods face challenges in efficiency and inter-observer variability.
Purpose of the Study:
- To validate a deep learning-based artificial intelligence (AI) tool for prostate cancer detection in biopsy samples.
- To assess the AI tool's capability in Gleason grading of prostate cancer.
- To evaluate the AI tool's performance across diverse external cohorts and scanning platforms.
Main Methods:
- Retrospective analysis of five external cohorts comprising 5922 H&E stained prostate biopsy sections (7473 cores from 423 cases).
- AI tool evaluated for tumor detection accuracy and Gleason grading performance against a panel of 11 expert urologic pathologists.
- Statistical validation using sensitivity, specificity, negative predictive value (NPV), and quadratically weighted kappa for grading agreement.
Main Results:
- PCA detection classifier demonstrated high accuracy: sensitivity (0.971-1.000), specificity (0.875-0.976), and NPV (0.988-1.000) across cohorts.
- AI tool identified tumor tissue missed by pathologists in several cases; false positives mainly involved suspicious lesions or mimics.
- Gleason grading agreement for the AI tool was comparable to experienced pathologists (kappa 0.77/0.72 single, 0.903/0.855 consensus).
Conclusions:
- The validated AI tool exhibits high accuracy for prostate cancer detection in biopsy specimens, independent of institute or scanner.
- AI-driven Gleason grading performance is comparable to that of expert genitourinary pathologists.
- This deep learning classifier shows significant potential to enhance efficiency and consistency in prostate biopsy pathology workflows.
