Foundation Model and Multi-Instance Learning-Based Framework for Predicting Lymphovascular Invasion in Prostate...

Area of Science:

• Pathology
• Artificial Intelligence
• Genomics

Background:

• Lymphovascular invasion (LVI) is a critical prognostic factor in prostate cancer (PCa).
• Accurate prediction of LVI is essential for patient management.
• Current methods for LVI detection can be challenging.

Purpose of the Study:

• To develop and validate an AI-based framework for predicting LVI in PCa.
• Leverage multi-instance learning (MIL) and foundation models for WSI analysis.
• Ensure accurate and interpretable LVI prediction.

Main Methods:

• Implemented a weakly supervised deep-learning pipeline using clustering-constrained attention MIL.
• Analyzed H&E-stained WSIs from two independent cohorts (RHWU and TCGA).
• Utilized pretrained encoders (UNI-v2, CONCH, ResNet-50) and attention heatmaps for interpretability.

Main Results:

• Achieved strong predictive performance, with UNI-v2 yielding AUCs of 0.839 (RHWU) and 0.854 (TCGA).
• Attention heatmaps identified high-risk regions with specific histopathologic features.
• Transcriptomic analysis revealed DEGs linked to mitotic and immune pathways in LVI-positive cases.

Conclusions:

• Developed a robust and interpretable AI framework for LVI prediction in PCa from WSIs.
• The AI model demonstrated high accuracy and provided biologically meaningful insights.
• The framework shows potential for clinical translation as a decision-support tool in precision pathology.