Multi-View Stenosis Classification Leveraging Transformer-Based Multiple-Instance Learning Using Real-World Clinical Data

2026-02-02

AI Summary

This work addresses the challenge that existing single-view deep learning models for coronary angiography rely on costly view-level annotations and struggle to effectively capture the temporal and structural relationships across multiple angiographic views. To overcome this limitation, we propose SegmentMIL, a Transformer-based multiple instance learning framework that, for the first time, enables joint multi-view modeling using only patient-level weak labels from real-world clinical data. Without requiring view-level supervision, SegmentMIL accurately distinguishes and localizes the left and right coronary arteries along with their anatomical segments, while simultaneously performing stenosis classification and regional localization. Experimental results demonstrate that SegmentMIL significantly outperforms both single-view models and conventional MIL baselines in both internal and external evaluations, highlighting its strong clinical scalability and deployment potential.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Segmentation
• Multiagent Systems: Multiagent Learning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Security and Privacy: Data transparency and provenance

Abstract

Coronary artery stenosis is a leading cause of cardiovascular disease, diagnosed by analyzing the coronary arteries from multiple angiography views. Although numerous deep-learning models have been proposed for stenosis detection from a single angiography view, their performance heavily relies on expensive view-level annotations, which are often not readily available in hospital systems. Moreover, these models fail to capture the temporal dynamics and dependencies among multiple views, which are crucial for clinical diagnosis. To address this, we propose SegmentMIL, a transformer-based multi-view multiple-instance learning framework for patient-level stenosis classification. Trained on a real-world clinical dataset, using patient-level supervision and without any view-level annotations, SegmentMIL jointly predicts the presence of stenosis and localizes the affected anatomical region, distinguishing between the right and left coronary arteries and their respective segments. SegmentMIL obtains high performance on internal and external evaluations and outperforms both view-level models and classical MIL baselines, underscoring its potential as a clinically viable and scalable solution for coronary stenosis diagnosis. Our code is available at https://github.com/NikolaCenic/mil-stenosis.

Problem

• coronary artery stenosis
• multi-view angiography
• view-level annotations
• temporal dynamics
• clinical diagnosis

Innovation

• Transformer-based MIL
• multi-view learning
• weakly supervised learning
• coronary stenosis classification
• clinical vision