LST

LST performs automated segmentation of T2-hyperintense lesions on FLAIR MRI to detect and quantify brain lesions for studies including multiple sclerosis, diabetes mellitus, and Alzheimer's disease.

Key Features:

• Automated Lesion Detection: Employs an algorithm for automated detection of T2-hyperintense lesions in FLAIR images to reduce user bias.
• Multi-Modal Imaging Integration: Requires a three-dimensional (3D) gradient echo (GRE) T1-weighted image alongside a FLAIR image (acquired at 3 Tesla) and classifies tissue into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF).
• Iterative Outlier Analysis: Analyzes FLAIR intensity distributions within each tissue class to identify outliers indicative of lesions and iteratively expands lesion labels from conservative to liberal interpretations using voxel-by-voxel analysis.
• Likelihood-Based Assignment: Computes likelihoods of voxels belonging to WM or GM versus being part of lesions and assigns lesion probability probabilistically to refine segmentation.

Scientific Applications:

• Multiple Sclerosis: Automated detection and quantification of T2-hyperintense white matter lesions for MS diagnosis and prognosis prediction.
• Other Neurological Conditions: Segmentation of brain lesions associated with diabetes mellitus and Alzheimer's disease for comparative lesion burden analysis.
• Research and Clinical Trials: Provides quantitative lesion measures suitable for basic research and clinical trial endpoints.

Methodology:

Performs tissue classification using 3D GRE T1-weighted and FLAIR images, analyzes FLAIR intensity distributions per tissue class to detect outliers, applies voxel-by-voxel iterative expansion and likelihood-based assignment comparing WM/GM probability versus lesion probability, and was validated on scans from 53 MS patients, 10 patients with posterior fossa lesions, and 18 controls on a Philips Achieva 3T scanner achieving R2 values >0.93 across FLAIR slice thicknesses up to 6 mm.