Evaluation of a deep learning segmentation tool to help detect spinal cord lesions from combined T2 and STIR acquisitions in people with multiple sclerosis - Report - MDSpire
Advertisement
Evaluation of a deep learning segmentation tool to help detect spinal cord lesions from combined T2 and STIR acquisitions in people with multiple sclerosis
Deep Learning Segmentation Enhances Spinal Cord Lesion Detection in MS Using T2 and STIR MRI
Overview
A deep learning (DL) segmentation tool combining T2 and STIR spinal cord MRI sequences significantly improved clinicians' sensitivity in detecting multiple sclerosis (MS) lesions. The tool also reduced interpretation time and decreased inter-reader variability among radiologists and neurologists.
Background
Spinal cord lesions are present in up to 80% of people with multiple sclerosis and are critical for diagnosis and prognosis. MRI of the spinal cord is routinely performed but challenging due to small tissue size and artifacts. International guidelines recommend acquiring at least two MRI sequences, such as T2-weighted and STIR, to improve lesion detection. While DL models have been applied to brain MRI in MS, their use for spinal cord lesion detection, especially combining multiple sequences, has not been clinically evaluated until now.
Data Highlights
Metric
Value
Training Dataset
140 cases from 40 scanners
Test Dataset
40 cases
Lesion-wise Sensitivity (Model)
0.89
Lesion-wise Precision (Model)
0.64
Study Cohort
50 patients with MS
Readers
20 clinicians (radiologists and neurologists)
Key Findings
The DL tool achieved a lesion-wise sensitivity of 0.89 and precision of 0.64 on a test set.
Use of the tool increased clinicians' sensitivity to detect spinal cord MS lesions compared to unaided reading.
The tool reduced interpretation time per MRI volume, enhancing workflow efficiency.
Inter-reader variability decreased when clinicians used the DL segmentation aid.
The study included a diverse group of readers with varying expertise, demonstrating broad applicability.
Ground truth was established by expert consensus using multiple MRI timepoints and sequences.
Clinical Implications
Incorporating a DL segmentation tool that analyzes combined T2 and STIR spinal cord MRI sequences can improve lesion detection sensitivity and reduce reading time in clinical practice. This approach supports more consistent and accurate diagnosis and monitoring of MS spinal cord involvement, potentially leading to better patient management.
Conclusion
The evaluated DL segmentation method effectively aids clinicians in detecting spinal cord lesions in MS using combined T2 and STIR imaging, improving sensitivity and efficiency. This tool represents a promising advancement for routine clinical MRI interpretation in MS care.
