Segmentation of White Matter Lesions from Volumetric MR Images
Quantitative analysis of the changes to the brain’s white matter is an important objective for a better understanding of pathological changes in various forms of degenerative brain diseases. To achieve an accurate quantification, an algorithm is proposed for automatic segmentation of white matter atrophies and lesions from T1-weighted 3D Magnetic Resonance (MR) images of the head. Firstly, white matter, gray matter and cerebrospinal fluid (CSF) compartments are segmented. Then, external and internal cisterns are separated by placing cutting planes relative to the position of the anterior and posterior commissure. Finally, a region growing method is applied to detect lesions inside the white matter. Since lesions may be adjacent to the gray matter, we use the external cisterns as a clue to prevent the algorithm from absorbing low gray level points in the gray matter.
The method is fully applied to detect the white matter lesions and relevant structures from a set of 41 MR images of normal and pathological subjects. Subjective assessment of the results demonstrates a high performance and reliability of this method.