Co-segmentation of MR and MR spectroscopy imaging using hidden markov models

A Hidden Markov Models based technique is introduced for co-segmentation of MRI and MRSI data of the brain. The technique demonstrates the ability of Hidden Markov Models to handle the co-analysis of MRI and MRSI for the purpose of improving the accuracy of MRI segmentation as well as the quantification of brain metabolites. For that purpose, two HMM-based schemes are presented; one that relies on parallel HMMs for separately analyzing MRI and MRSI data and the other utilizes combined feature vectors of MRI and MRSI data. The co-segmentation of MRI and MRSI data using HMMs is evaluated using simulated MRI brain data (from the McConnell Brain Imaging Centre, Montreal Neurological Institute of McGill University) and simulated MRSI data. Experimental results demonstrate that the co-segmentation of brain MRI and MRSI data based on HMMs exhibited higher accuracy, in terms of the Dice similarity coefficient, than only using brain MRI data. The technique involving parallel HMMs that separately analyze brain MRI and MRSI data and then combine the segmentation results demonstrated better accuracy and faster segmentation times compared to the co-analysis of combined MRI and MRSI data of the brain.