Development of a deep learning system for predicting biochemical recurrence in prostate cancer
Lu Cao1, Ruimin He1, Ao Zhang2
1Department of Pathology, Tianjin Medical University Cancer Institute and Hospital, National Clinical Research Center for Cancer, Key Laboratory of Cancer Prevention and Therapy, Tianjin's Clinical Research Center for Cancer, Key Laboratory of Breast Cancer Prevention and Therapy, Tianjin Medical University, Ministry of Education, Tianjin, 300060, China.
BMC Cancer
|February 10, 2025
Summary
A new deep learning system accurately predicts biochemical recurrence (BCR) in prostate cancer (PCa) patients using biopsy images. This AI-driven approach aids in developing targeted prostatectomy strategies for better patient outcomes.
Area of Science:
- Oncology
- Pathology
- Artificial Intelligence
Background:
- Biochemical recurrence (BCR) affects 20%-40% of men post-prostatectomy for prostate cancer (PCa).
- Current prediction methods, like Gleason grading, miss crucial histopathological details, limiting prognostic accuracy.
- There's a need for advanced methods to predict BCR risk more effectively before surgery.
Purpose of the Study:
- To develop and validate a deep learning system for predicting PCa BCR using prostate biopsy images.
- To assess the system's performance and clinical utility in preoperative risk stratification.
- To explore the interpretability of AI-generated features in relation to pathological findings.
Main Methods:
- Utilized 1585 prostate biopsy images from 317 patients.
- Employed the Inception_v3 neural network for patch-level image analysis.
- Applied multiple instance learning for whole slide image feature extraction.
- Integrated deep learning features with machine learning algorithms for patient-level prediction.
Main Results:
- The AI system achieved high performance in predicting BCR (AUC = 0.911).
- Decision Curve Analyses indicated potential for favorable clinical benefits.
- Performance improved with an increased number of whole slide images (WSIs) per patient.
- Demonstrated correlation between AI features and pathological findings, suggesting model interpretability.
Conclusions:
- Deep learning models can accurately predict BCR risk from prostate biopsy samples.
- This AI-driven prediction facilitates the formulation of targeted treatment strategies for PCa.
- The system offers a promising tool for improving preoperative risk assessment and patient management.
