An Architecture for the Recognition and Classification of Multiple Sclerosis Lesions in Mr Images

A software architecture is presented that is able to perform classification of focal lesions due to the multiple sclerosis disease in MR images of the brain. The methodology proceeds through four main steps: tissue segmentation, re-clustering and tissue classification, lesion localization and lesion classification. Images are first segmented using the FCM algorithm; then the images of each cluster are processed in order to classify and label non-pathologic tissues making use of simple decision algorithms based on suitable numerical indices related to tissue morphology. All possible candidates to be sclerosis lesions are then located by means of morphological operations applied to binary images of single tissues. Finally the classification step is performed together with an estimate of the position and the shape for each lesion. Each candidate has been characterized by means of a set of measurements related to its shape, position and brightness as a way to code the clinical knowledge about the disease under investigation. Classification is implemented using both an algorithmic classifier and a multi-layer perceptron trained using a back-propagation scheme, and the performances of the two approaches are compared. The outline of the whole architecture is presented and the experimental results are reported.