A fuzzy based adaptive BPNN learning algorithm for segmentation of the brain MR images

Segmentation is an important step in the processing of MR images for the purpose of medical diagnosis, 3-D visualization of the human brain. It is a very difficult problem to segment multiple tissues in a single channel MR image. In this work the features of the three standard MR images i.e. T1, T2 and PD weighted images have been employed that has not only improved the accuracy of the segmentation process but also enhanced its reliability. The supervised BPNN has been used for the classification of the feature vectors in this work. The fuzzy based adaptive control strategy has been used for the first time in the multiple segmentation problems that has shown tremendous effect on the learning efficiency of the BPNN. In order to improve the partial volume effect the four neighboring pixels from each standard image have been utilized. For the removal of the extra cranial parts of the brain, a new and reliable morphological method has been employed. The results of the segmentation have been compared with the radiologist marked ground truth.