Multi Atlas Based Segmentation
In collaboration with
Multi Atlas Based Segmentation (MABS) is a methodology for automatic image segmentation that is proving to be able to provides robust and reliable results.
The rationale behind MABS is to use a set of digital atlases (anatomical image plus labels) and image registration technique to “bring” the the contours from the atlases to the query image. Applying a statistical methodology to all the deformed labels the final segmentation will be extracted. Of course the number and the “type” of the atlases that will be used in this process can significantly influence the quality of the final results and also the time required to get them. In fact it would be advantageous to not include in the workflow the atlases that introduce redundant or noisy information. This filtering process is usually called “atlas selection”.
In our work we are implementing a MABS algorithm into the Plastimatch project (www.plastimatch.org). In this software have been introduced many interesting and unique features in order to solve scientific and practical issues related to this type of algorithms. Complete and smart workflow that allows to use data directly from a clinical enviroment, atlas selection based on different strategies, multiple ways for final contour extracting (gaussian and/or staple), fast and accurate image registration process are the main advantages of our implementation.
Several tests are running on head and neck CT dataset (radiotherapy purpose) and brain MRI dataset (neuroscience application) to validate the proposed metodology and its innovative features.