Automated white matter total lesion volume segmentation in diabetes

Area of Science:

• Neuroimaging
• Medical image analysis
• Diabetology

Background:

• White matter (WM) lesion segmentation is crucial for assessing neurological conditions.
• Manual segmentation is time-consuming, leading to reliance on subjective rating scales.
• Automated methods offer a more efficient and objective approach.

Purpose of the Study:

• To compare the performance of an automated WM lesion segmentation algorithm with subjective rating scales and expert manual segmentation.
• To validate the Lesion Segmentation Toolbox (LST) for total lesion volume assessment in a cohort of patients with type 2 diabetes.

Main Methods:

• Structural T1 and FLAIR MRI data from 50 diabetic and 50 non-diabetic subjects were analyzed.
• WM lesion segmentation and total lesion volume were generated using the Statistical Parametric Mapping (SPM8) Lesion Segmentation Toolbox.
• Subjective WM lesion grading was performed by two readers using a 0-9 rating scale; manual segmentation served as the ground truth.

Main Results:

• High correlation (ρ = 0.87) was found between the Lesion Segmentation Toolbox and ground-truth manual segmentation at a threshold of k=0.25.
• Correlation between subjective lesion grading and the Lesion Segmentation Toolbox was lower (ρ = 0.73 at k=0.15).
• Subjective raters may overestimate white matter lesion burden compared to automated methods.

Conclusions:

• The Lesion Segmentation Toolbox is validated for determining total lesion volume in diabetes-enriched populations.
• The LST provides a readily available and objective substitute for subjective WM lesion scoring.
• This automated method is suitable for studies involving diabetes and other conditions with leukoaraiosis.