M-SCAN: A Multistage Framework for Lumbar Spinal Canal Stenosis Grading Using Multi-View Cross Attention

2025-03-03

AI Summary

Lumbar spinal stenosis (LSS) diagnosis via MRI suffers from high inter-reader variability and substantial radiologist workload. To address these challenges, we propose the first multi-stage, cross-view deep learning framework for automatic LSS grading, jointly processing axial T2-weighted and sagittal T1- and T2-STIR sequences. Our method introduces a novel cross-view feature alignment mechanism and a sequential multi-view Transformer architecture to overcome limitations of single-view modeling. Leveraging stage-wise supervised training and 3D volumetric representation, the model achieves an AUROC of 0.971 on a real-world clinical dataset of 1,975 cases—significantly outperforming existing approaches. It demonstrates strong robustness, end-to-end automation, and inherent clinical interpretability. This work establishes a new paradigm for standardized, objective LSS diagnosis in routine clinical practice.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

The increasing prevalence of lumbar spinal canal stenosis has resulted in a surge of MRI (Magnetic Resonance Imaging), leading to labor-intensive interpretation and significant inter-reader variability, even among expert radiologists. This paper introduces a novel and efficient deep-learning framework that fully automates the grading of lumbar spinal canal stenosis. We demonstrate state-of-the-art performance in grading spinal canal stenosis on a dataset of 1,975 unique studies, each containing three distinct types of 3D cross-sectional spine images: Axial T2, Sagittal T1, and Sagittal T2/STIR. Employing a distinctive training strategy, our proposed multistage approach effectively integrates sagittal and axial images. This strategy employs a multi-view model with a sequence-based architecture, optimizing feature extraction and cross-view alignment to achieve an AUROC (Area Under the Receiver Operating Characteristic Curve) of 0.971 in spinal canal stenosis grading surpassing other state-of-the-art methods.

Problem

• Automates lumbar spinal canal stenosis grading using deep learning.
• Reduces labor-intensive MRI interpretation and inter-reader variability.
• Integrates multi-view MRI images for improved diagnostic accuracy.

Innovation

• Multistage deep-learning framework automates stenosis grading
• Multi-view model integrates sagittal and axial MRI images
• Sequence-based architecture optimizes feature extraction and alignment