Curtailing Heterogeneous Disparity in Brain Imaging and Dna Structure for Pattern Recognition Based Estimation of Alzheimer’S Disease

Alzheimer’s disease is a multifactorial and progressive neurodegenerative disorder that affects an individual’s memory and cognitive skills. It is a major cause of death around the globe and according to 2015 Alzheimer’s association report, the death percentage has increased to 71% since year 2000. The clinical symptoms of the disease appear at a stage when the loss has become irreversible. Modern brain imaging techniques have enabled us to non-invasively visualize the internal structures of the brain. Scientists believe that structural and functional changes due to Alzheimer’s disease begin in the brain more than 20 years before any clinical symptoms are observed. Early detection of the disease is crucial for the patient, care givers and relatives to cope with the situation. It will also help medical practitioners to discover new drugs. For this reason there is an imperative need of image based automated techniques to assist medical professionals in correct diagnosis of Alzheimer’s disease using brain images. In recent years, there is an intensive research focused on the identification of Alzheimer’s disease from brain images using machine learning methods. Structural brain images like MRI have been extensively used in this regard. In our research work, we have proposed an automated image processing based approach for the early identification of Alzheimer’s disease from MRI scans of the brain. The dataset selected consists of 236 age and gender matched individuals and the features selected are volume of GM, WM and CSF, and size of hippocampus. In addition to image features, genetic aspects of Alzheimer’s disease are also considered in classification task. Well known APOE risk gene data and 14 SNP data associated with Alzheimer’s disease are incorporated in the feature set. Seven different classification models from different algorithmic paradigms are used for identification of patients and controls. For evaluation of our scheme, we have used cross validation and 66% vi ii split test strategy. Classification results are obtained using image features, genetic features and combination of the both. It is observed that image features produced best classification of cases and controls. On the other hand, genetic data can be very useful in predicting the risk of disease well before any changes to brain are observed. The proposed approach is novel because it has been able to achieve higher accuracy/specificity/sensitivity values even using smaller feature set which is not the case of existing approaches. Moreover, the proposed approach is capable of identifying AD patients in early stages. Best results (100% accuracy, 100% specificity, 100% sensitivity) are achieved using volume of GM and size of left hippocampus with J48 classifier. Similarly APOE risk gene predicted the disease with 75% accuracy for all classifiers whereas SNP data achieved 86% accuracy with Naïve Bayes and SVM. The proposed approach will play a vital role in the domain of Computer Aided Diagnostics and Preventive Studies.