A ROBUST OPTIMIZED FEATURE SET BASED AUTOMATIC CLASSIFICATION OF ALZHEIMER’S DISEASE FROM BRAIN MR IMAGES USING K-NN AND ADABOOST

Abstract

For individuals suffering from some cognitive impairment, treatment plans will be greatly help patients and medical practitioners, if early and accurate detection of Alzheimer's disease (AD) is carried out. Brain MR Scans of patients' with health history and supportive medical tests results can lead to distinguish between Healthy/ Normal Controls (NC), Mild Cognitive Impairment (MCI) and AD patients. However manual techniques for disease detection are labour intensive and time consuming. This work is towards the development of Computer Aided Diagnosis (CAD) tool for Alzheimer's disease detection and its classification into the early stage of AD i.e. MCI and later stage -AD. The paper is about selection of robust optimized feature set using combination of forward selection and/or backward elimination method with K-NN classifier and validation of results with features selected (using forward selection method); with Ada-boost for improved classification accuracy. The features are extracted on Gray Level Cooccurrence Matrix (GLCM). The experimentation is based on Public Brain Magnetic Resonance datasets named Open Access Series of Imaging Studies (OASIS) [7] with patients diagnosed with NC, MCI and AD. The four models considered for automatic classification arei. Abnormal vs. Normal; ii. AD vs. MCI; iii. MCI vs. NC and iv. AD vs. NC. Feature set optimized using K-NN and validated with AdaBoost has given improved classification accuracy for each model. The output of developed CAD system is compared with Radiologists opinion for test images and has shown 100% match between the output of computer algorithm and experts opinion for some important models under consideration.