Microstructure-informed deep learning improves thalamic atrophy segmentation and clinical associations in multiple sclerosis and related neuroimmunological diseases - PubMed
4 days ago
- #deep learning
- #multiple sclerosis
- #thalamic atrophy
- Microstructure-informed deep learning improves thalamic atrophy segmentation in multiple sclerosis (MS) and related neuroimmunological diseases.
- Thalamic atrophy is a sensitive imaging marker of neurodegeneration in MS and related disorders.
- Four algorithms (FreeSurfer, FIRST, DBSegment, MindGlide) were benchmarked against ground truth (GT) labels for thalamus segmentation.
- MindGlide, an MS-trained deep learning model, balanced sensitivity and precision best among tested methods.
- Quantitative R1 maps provided modest or no improvement in segmentation accuracy when added to MindGlide.
- MindGlide volumes showed the strongest associations with disability and cognitive scores cross-sectionally.
- Longitudinally, MindGlide detected the largest effects between thalamic volume change and EDSS worsening.
- Higher-resolution qMRI and multi-contrast deep learning may further enhance thalamic segmentation in neuroinflammatory diseases.