The objective of the proposed statistical modeling scheme is to automatically detect, segment and track-in-time regions pertaining to relapsing-remitting multiple sclerosis (MS) lesions in MR image-sequences of the brain. Unsupervised clustering via Gaussian mixture modeling is utilized to extract coherent space-time regions (space-time “blobs”) in a four-dimensional feature space (intensity, position (x,y), and time) and corresponding coherent segments in the sequence content. The parameters of the model are determined via the Expectation-Maximization (EM) algorithm according to the maximum likelihood principle. In the detection stage of the framework, the regions corresponding to MS lesions are identified out of the collection of extracted segments based on region-level rules. Following the detection, lesion segmentation and tracking are performed in a unified manner via global space-time modeling.
The performance of the proposed framework was demonstrated using a well-known benchmark sequence of 24 MR images, acquired from a relapsing-remitting MS patient over a period of approximately a year. Strong correlation was achieved between the automatically obtained segmentation results and an expert’s manual segmentation. Robustness tests via a simulation tool further confirmed the validity of the suggested methodology.
Examples of Lesion Segmentation and Tracking in Time
For a detailed explanation read the Thesis.
For problems or
questions regarding this web page contact