A new shape based techique for classification and registration : application to multimodal medical images

The present paper is part of an investigation being carried out by the autors for developing a knowledge based framework for combining different modalities of medical imaging such as CT, MRI, PET, SPECT, USG etc. which ever is relevant for a particular case for medical diagnostics and therapeutic planning. In this paper we are mainly concerned about some brain diseases to study the degree of prognosis of the disease. The study consists of two parts (a) localisation and slassification of intracranial tumors in central nervous system (CNS) appering in brain as space occupying lesion and (b) registration of CT, [T1] and [T2] weighted MR images of brain having Alzheimer's disease. Both the classification and the registration technique are based on mathematical theory of shape briefly presented here. In first part of the paper the brain tumors are clssified in noncancerous benign group and in cancerous malignant group implementing the concept of "shape distance" and "shape similarity measure" as defined by D. Dutta Majumder et al. [2],[3],[4],[5],[6],[7],[8],[9],[10],[24]. A finer gradation is suggested regarding the growth of the tumors in brain to assist the clinicians to take necssary decision for proper surgical planning. In second part of the paper the registration of region of interest (ROI) of brain is performed for Alzheimer's patient considering CT, [T1] and [T2] weighted MR images.