Detection of subtle white matter lesions in MRI through texture feature extraction and boundary delineation using an...

Area of Science:

• Neuroimaging
• Medical image analysis
• Computational neuroscience

Background:

• White matter lesions (WML) are indicative of various brain disorders.
• Accurate segmentation of WML is vital for tracking disease progression and evaluating treatment efficacy.
• Detecting subtle, early-stage WML remains a significant challenge in automated segmentation.

Purpose of the Study:

• To develop and validate an automated approach for segmenting mild white matter lesion loads.
• To improve the accuracy of detecting subtle WML, particularly in early disease stages.

Main Methods:

• Utilized an intensity standardization technique.
• Employed a Gray Level Co-occurrence Matrix (GLCM) embedded clustering technique for feature extraction.
• Integrated a Random Forest (RF) classifier for morphology identification.
• Applied a Local Outlier Factor (LOF) algorithm to precisely define lesion boundaries by identifying edge pixels based on local density deviations.

Main Results:

• The automated approach demonstrated strong agreement and correlation with manual segmentation by a neuroradiologist (ICC = 0.881, Pearson r = 0.895).
• The method outperformed three leading algorithms in five out of six key metrics from the MICCAI Grand Challenge.
• Validation on 32 human subjects confirmed the robustness and accuracy of the WML segmentation.

Conclusions:

• The proposed automated WML segmentation method effectively identifies subtle lesions.
• This technique facilitates more accurate segmentation, potentially enabling earlier diagnosis and intervention for brain disorders.
• The approach shows promise for improving the evaluation of disease course and therapeutic interventions in clinical drug discovery.