Automated Segmentation of White Matter Lesions Using Global Neighbourhood Given Contrast Feature-Based Random Forest and Markov Random Field

Recent studies show that, cerebral White Matter Lesion (WML) is related to cerebrovascular diseases, cardiovascular diseases, dementia and psychiatric disorders. Manual segmentation of WML is not appropriate for long term longitudinal studies because it is time consuming and it shows high intra-and inter-rater variability. In this paper, a fully automated segmentation method is utilized to segment WML from brain Magnetic Resonance Imaging (MRI). The segmentation method uses a combination of global neighbourhood given contrast feature-based Random Forest (RF) classifier and Markov Random Field (MRF) to segment WML. To remove false positive lesions we use a rule based morphological postprocessing operation. Quantitative evaluation of the proposed method was performed on 24 subjects of ENVIS-ion study. The segmentation results were validated against the manual segmentation, performed by an experienced radiologist and were compared to a recently published WML segmentation method. The results show a dice similarity index of 0.75 for high lesion load, 0.71 for medium lesion load and 0.60 for low lesion load.