Related Experiment Video
Updated: Aug 23, 2026
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
Whole-brain atrophy in multiple sclerosis measured by automated versus semiautomated MR imaging segmentation
Jitendra Sharma1, Michael P Sanfilipo, Ralph H B Benedict
1Buffalo Neuroimaging Analysis Center, University at Buffalo, State University of New York, Buffalo, NY, USA.
Background And Purpose:
Semiautomated and automated methods are used to measure whole-brain atrophy in multiple sclerosis (MS), but their comparative reliability, sensitivity, and validity are unknown.
Methods:
Brain parenchymal fraction (BPF) was measured in patients with MS (n = 52) and healthy control subjects (n = 17) by four methods: semiautomated or automated segmentation and 2D or 3D pulse sequences. Linear measures of atrophy, whole-brain lesion volumes, and clinical data were used to explore validity.
Results:
The 2D automated method yielded unreliable segmentation and was discarded. The three other BPF methods produced data that were highly intercorrelated and indistinguishable by analysis of variance. In the MS group, semiautomated (2D: 0.84 +/- 0.04, P <.001; 3D: 0.84 +/- 0.05, P =.04) and automated 3D (0.83 +/- 0.05, P =.002) BPFs were lower than controls (semiautomated 2D: 0.88 +/- 0.02; 3D: 0.88 +/- 0.03; automated 3D: 0.88 +/- 0.03). In the MS group, the semiautomated (r = -.79 to -.82) and automated 3D (r = -.81) BPFs inversely correlated with third ventricular width and showed similarly robust correlations with the bicaudate ratio (all r = -.74). The semiautomated and automated BPFs showed similar, moderate correlations with T1 hypointense and FLAIR hyperintense lesion volume, physical disability (Expanded Disability Status Scale) score, and disease duration and similar differences between secondary progressive and relapsing-remitting patients. The intraobserver, interobserver, and test-retest reliability was somewhat higher for the automated than for the semiautomated methods.
Conclusion:
These automated and semiautomated measures of whole-brain atrophy provided similar and nearly interchangeable data regarding MS. They discriminated MS from healthy individuals and showed similar relationships to established disease variables.
Insights
Automated and semiautomated methods for measuring whole-brain atrophy in multiple sclerosis (MS) provide similar, interchangeable data. These techniques effectively distinguish MS patients from healthy individuals and correlate with disease severity.
Area of Science:
- Neuroimaging
- Medical Physics
- Radiology
Background:
- Whole-brain atrophy measurement in multiple sclerosis (MS) is crucial for disease monitoring.
- Semiautomated and automated imaging techniques are increasingly used, but their comparative performance is not well-established.
Purpose of the Study:
- To compare the reliability, sensitivity, and validity of semiautomated and automated methods for measuring brain parenchymal fraction (BPF) in MS.
- To assess the correlation of these BPF measures with established clinical and imaging markers of MS severity.
Main Methods:
- Brain parenchymal fraction (BPF) was measured in 52 MS patients and 17 healthy controls using four methods: semiautomated/automated segmentation with 2D/3D pulse sequences.
- Validity was explored using linear atrophy measures, lesion volumes, and clinical data.
Main Results:
- The 2D automated method was unreliable and excluded. The remaining three BPF methods yielded highly correlated and statistically indistinguishable results.
- All three methods showed significantly lower BPF in MS patients compared to controls.
- BPF measures demonstrated strong inverse correlations with ventricular width and bicaudate ratio, and moderate correlations with lesion load and disability scores.
Conclusions:
- Semiautomated and automated BPF measurements offer similar, nearly interchangeable data for assessing whole-brain atrophy in MS.
- These methods reliably differentiate MS patients from healthy individuals and correlate consistently with key disease variables.
