Identification of fronto temporal Dementia in brain mri using Artificial intelligence techniques

Abstract

Atrophy of frontal and temporal lobes in the human brain causes newlinedysfunctionality in Language, Emotion, Executive abilities. These losses are newlineirreversible. If atrophy is detected at an early age it is beneficial for the patient. newlineA detailed diagnosis of the patient in terms of physical, physiological, newlinepsychological and neuropsychological aspects is mandatory to trace the newlinepossibility of FTD. For the study, MR imaging modality is used. newlineThere is a continuous increase in the number of patients affected by the newlinedisease but a limited number of specialists (neurologist, neuro-radiologist) are newlineavailable in most of the nations across the world. This deficit greatly affects newlinethe patients who need utmost and immediate care. Hence there is a need for newlinethe system which helps the doctors in diagnosing the disease accurately. As newlinethe degree of atrophy or deterioration varies from patient to patient, it is very newlinedifficult to exactly identify FTD easily. Sometimes the manual diagnosis can newlinebe error-prone also. Using the proposed method images are classified and give newlineinformation about the severity of deterioration. An attempt is also made to newlinecalculate the area of deterioration (atrophy) region. newlineThe method started with a channel selection providing better contrast. The newlinecontrast of the image was enhanced using CLAHE. Segmentation was newlineachieved using Otsu s method and Mathematical Morphology. Using newlineClustering methods like FCM, K-Means an attempt was made to calculate the newlinearea of demented regions. newlineImportant textural statistics were obtained using Gray Level Co-occurrence newlineMatrix (GLCM) were fed into the multilayer feed-forward network for newlineclassification. For the training process, back propagation ANN with a sigmoid newlineneuron in the hidden layer and a linear neuron in the output layer was used. In newlinethe Feed Forward BPN, extracted GLCM features were fed to the input layer newlineand the error signal was computed. The error was calculated by comparing the newlineactual output obtained (output of the network) with the targeted output newline(expected output). The error