Stroke Lesion Segmentation in FLAIR MRI Datasets Using Customized Markov Random Fields.

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Radiology

Background:

• Accurate segmentation of stroke lesions is vital for using lesion volume as a clinical trial endpoint.
• Current segmentation methods often require multi-modal Magnetic Resonance Imaging (MRI) data.

Purpose of the Study:

• To develop and evaluate a novel method for segmenting sub-acute ischemic stroke lesions using only Fluid-Attenuated Inversion Recovery (FLAIR) MRI.
• To assess the robustness and reliability of the proposed segmentation technique.

Main Methods:

• A Bayesian technique utilizing Gabor textures from FLAIR signal intensities for initial segmentation.
• A customized voxel-level Markov Random Field model incorporating intensity and Gabor texture features for refinement.
• Validation on 151 multi-center datasets using a leave-one-patient-out approach.

Main Results:

• Achieved an average Dice coefficient of 0.582, competitive with multi-modal MRI methods.
• Demonstrated superior performance compared to leading Convolutional Neural Network and level-set methods from the ISLES 2015 challenge.
• The method provides robust lesion segmentation using only FLAIR MRI sequences.

Conclusions:

• The proposed method offers a reliable and robust approach for segmenting sub-acute ischemic stroke lesions from FLAIR MRI.
• This technique has the potential to serve as a valuable tool in clinical trials, particularly as a follow-up sequence analysis.