Benchmarking DINOv3 for Multi-Task Stroke Analysis on Non-Contrast CT

2025-09-27

AI Summary

Non-contrast CT (NCCT) suffers from low contrast and poor signal-to-noise ratio, severely impeding early and accurate stroke diagnosis. To address this, we propose the first application of the self-supervised vision Transformer DINOv3 to multi-task stroke analysis, establishing a unified framework that jointly performs infarct/hemorrhage segmentation, abnormality classification, hemorrhage subtype identification, and ASPECTS scoring. Our method leverages DINOv3 to extract robust high-level features and integrates them into a multi-task learning architecture. We comprehensively evaluate the framework across multiple public and private NCCT datasets. Results demonstrate substantial improvements in lesion detection and classification performance on low-quality NCCT scans, establishing a new state-of-the-art baseline. Furthermore, we publicly release our code and benchmark suite to accelerate clinical translation of self-supervised learning in neuroimaging.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Multi-instance/Multi-view Learning
• Search and Optimization: Non-convex Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Data transparency and provenance
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Non-contrast computed tomography (NCCT) is essential for rapid stroke diagnosis but is limited by low image contrast and signal to noise ratio. We address this challenge by leveraging DINOv3, a state-of-the-art self-supervised vision transformer, to generate powerful feature representations for a comprehensive set of stroke analysis tasks. Our evaluation encompasses infarct and hemorrhage segmentation, anomaly classification (normal vs. stroke and normal vs. infarct vs. hemorrhage), hemorrhage subtype classification (EDH, SDH, SAH, IPH, IVH), and dichotomized ASPECTS classification (<=6 vs. >6) on multiple public and private datasets. This study establishes strong benchmarks for these tasks and demonstrates the potential of advanced self-supervised models to improve automated stroke diagnosis from NCCT, providing a clear analysis of both the advantages and current constraints of the approach. The code is available at https://github.com/Zzz0251/DINOv3-stroke.

Problem

• Benchmarking DINOv3 for stroke analysis on NCCT
• Improving automated stroke diagnosis using self-supervised models
• Evaluating feature representations for multiple stroke classification tasks

Innovation

• Using DINOv3 for stroke analysis tasks
• Generating features with self-supervised transformer
• Establishing benchmarks for automated stroke diagnosis