Related Experiment Video
Updated: Jan 11, 2026
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
Clinical-Grade Interpretable Artificial Intelligence Tool for Automated Detection of Lymph Node Metastasis in
Fatemeh Zabihollahy1, Kangdi Shi2, Collins Wangulu3
1Laboratory Medicine Program, University Health Network, Toronto, Ontario, Canada; Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, Ontario, Canada; Division of Urology, Department of Surgery, University Health Network, Toronto, Ontario, Canada; Division of Surgical Oncology, Department of Surgery, University Health Network, Toronto, Ontario, Canada.
A novel artificial intelligence (AI) model accurately detects lymph node metastasis (LNM) in prostate cancer using limited data. This AI tool significantly reduces pathologist errors and speeds up slide examination, improving diagnostic efficiency.
Area of Science:
- Urology
- Pathology
- Artificial Intelligence
Background:
- Lymph node metastasis (LNM) is a critical prognostic factor in prostate cancer, influencing mortality and treatment strategies.
- Manual histopathological evaluation of LNM is time-consuming, variable, and error-prone.
- Deep learning for LNM detection requires extensive annotated datasets, which are often unavailable.
Purpose of the Study:
- To develop and validate a novel artificial intelligence (AI)-driven model for accurate and efficient detection of lymph node metastasis (LNM) in prostate cancer.
- To address the challenge of limited annotated data by incorporating informative instances from unlabeled data through iterative error correction.
- To assess the AI model's performance against human pathologists in terms of accuracy, miss rates, and examination time.
Main Methods:
- Developed an AI model using a limited annotated dataset and incorporating informative unlabeled data for iterative error correction.
- Validated the model on 787 whole slide images from 3 academic medical centers (>2000 lymph node tissues).
- Evaluated model performance using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and compared miss rates and slide examination times with pathologists.
Main Results:
- The AI model achieved an AUC of 0.94 (95% CI, 0.92-0.96) on a combined test set.
- Slide-level sensitivity and specificity were 96% and 92%, respectively.
- The AI model identified 17 micrometastases missed by pathologists, with a significantly lower miss rate (3% vs. 9%) and reduced slide examination time (P < .001).
Conclusions:
- The developed AI model offers a robust, autonomous, and reproducible method for accurate LNM detection in prostate cancer.
- The AI model demonstrates potential for clinical deployment, enhancing diagnostic accuracy and efficiency, particularly in identifying micrometastases.
- The AI approach effectively overcomes limitations of manual evaluation and data scarcity, showing interpretability in metastasis region identification.
