Regional Atrophy Analysis of Alzheimer Brain Magnetic Resonance Images Using Local Texture Patterns

Alzheimer Disease (AD) is the most common type of dementia among elderly people. It is a severe neurodegenerative disorder which is highly characterized by progressive loss of brain tissues. It interferes with normal activity of daily living due to loss of cognitive ability. Magnetic Resonance Imaging (MRI) has been proven to be very useful in early diagnosis and progression analysis of AD. This paper investigates the regional atrophy due to Alzheimer disease progression in four common brain tissues such as Cerebro Spinal fluid (CSF), Ventricle Segment (VS), White Matter (WM) and Gray Matter (GM) using their corresponding local texture patterns. The extracted information is used to classify Normal, Mild Cognitive Impaired (MCI) and AD subjects. An attempt is made to automatically segment the common brain tissues like WM, GM, CSF and VS. Features are extracted from their local texture patterns. Classification of Normal, MCI and AD is performed in order to investigate the efficiency of these extracted features as biomarkers in automated analysis of Alzheimer diagnosis. Anisotropic Diffusion filter based Level Set Method (ADLSM) is adapted to segment GM, WM, CSF and VS regions of brain. Fuzzy C means Clustering (FCM) technique is used to draw the initial contour which is later evolved using level set contour towards the desired boundaries. Local texture patterns such as Local Binary Patterns (LBP), Local Tetra Patterns (LTrP), Local Ternary Patterns (LTP) and Local Maximum Edge Binary Patterns (LMEBP) of segmented images are calculated. Histogram based features are extracted from these local patterns in order to classify NC, MCI and AD. It shows that the proposed FCM based ADLSM could able to segment the various brain tissues accurately. All local patterns are able to bring out the structural variations in terms of edge details. A maximum accuracy of 100 % is observed using LTP features and SVM classifier in differentiating AD and normal in GM, WM and whole brain regions. LMEBP features show an average performance measure of greater than 75% accuracy in differentiating MCI and normal subjects using SVM classifier. Similarly, LTP features show a good performance measure of 100% classification accuracy in differentiating AD and MCI in whole brain region. In conclusion, histogram features derived from local ternary patterns could be an efficient biomarker for classification of AD, MCI and normal subjects. Textural variations in gray matter and whole brain regions contribute more in differentiating the disease progression using local patterns. Hence, the proposed flow of segmentation algorithm, LTP feature extraction of various brain tissues along with SVM classifier may help to improve the automated diagnosis of Alzheimer disease progression.