Looking Where the Eye Cannot See: Improving MS Diagnosis with Deep Learning

2026-03-11 · Source: AI on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Medical Imaging AI · Depth: Intermediate, short

Summary

A new deep learning model, DeepMS, developed by CDS PhD student Jiajian Ma and colleagues, aims to reduce the high misdiagnosis rate of multiple sclerosis (MS). The model analyzes "normal-appearing white matter" (NAWM) in brain MRIs, areas that appear healthy to the human eye but may contain subtle microstructural injuries characteristic of MS. Diagnosing MS is challenging because its primary radiological signature, white matter lesions, is not unique to the disease. DeepMS was trained using a "joint dMRI–sMRI training" strategy, learning correlations between standard structural MRIs (sMRI) and specialized quantitative diffusion MRIs (dMRI). Once trained, DeepMS can operate solely on readily available sMRI scans, leveraging patterns learned from dMRI data. The model demonstrated diagnostic sensitivity comparable to "dissemination in space" (DIS) and specificity similar to the "central vein sign" (CVS), performing robustly across internal and 16 external datasets.

Key takeaway

For research scientists developing diagnostic AI, DeepMS demonstrates that integrating specialized imaging data during training can enable a model to extract subtle, clinically relevant information from more common, less specialized scans. This approach suggests that you should explore multimodal training strategies to enhance diagnostic sensitivity for diseases with ambiguous visual markers, even if the specialized modality is not available for routine inference. Consider open-sourcing your code to facilitate further research and validation.

Key insights

DeepMS improves MS diagnosis by analyzing subtle microstructural changes in normal-appearing white matter using deep learning.

Principles

• NAWM contains critical diagnostic signals beyond lesions.
• AI can extract diagnostic value invisible to human experts.

Method

DeepMS uses a joint dMRI–sMRI training strategy to learn correlations between standard structural MRIs and specialized diffusion MRIs, enabling inference from sMRI alone.

In practice

• Focus MS research on normal-appearing white matter.
• Explore multimodal training for subtle disease detection.

Topics

• Deep Learning
• Multiple Sclerosis Diagnosis
• Medical Imaging
• Diffusion MRI
• Normal-Appearing White Matter

Code references

• nyu-shenlab/DeepMS

Best for: Research Scientist, AI Researcher, AI Scientist, Data Scientist

Related on AIssential

• Learning neuroimaging models from health system-scale data — Health system-scale data enables AI foundation models to achieve superior, fair, and explainable neuroimaging diagnostics.
• 3D Classification of Paramagnetic Rim Lesions in Multiple Sclerosis via Asymmetric QSM-FLAIR Modeling — Asymmetric multimodal deep learning improves 3D classification of MS paramagnetic rim lesions using QSM and FLAIR.
• Vision-Language Model-Guided Deep Unrolling Enables Personalized, Fast MRI — PASS uses a VLM to guide personalized MRI sampling and reconstruction for faster, anomaly-aware imaging.
• REVEAL: Multimodal Vision-Language Alignment of Retinal Morphometry and Clinical Risks for Incident AD and Dementia Prediction — Jointly modeling retinal images and clinical risk factors via VLM improves early AD and dementia prediction.
• Enhancing Pathological VLMs with Cross-scale Reasoning — Explicit cross-scale reasoning training and a leakage-aware VQA benchmark enhance pathological VLM understanding across magnifications.
• Validating a Deep Learning Algorithm to Identify Patients with Glaucoma using Systemic Electronic Health Records — An EHR-only deep learning model can effectively pre-screen for glaucoma without specialized imaging.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI on Medium.