A Hybrid Cross Entropy Thresholding for Early Alzheimer's Disease Detection

Abstract

Alzheimer's disease is an advanced form of brain disorder that touches a person’ ability and disallows him to perform his daily tasks. Early diagnosis in this disease is important because it allows treatments that slow down its progression. A diagnostic indicator in patients with Alzheimer disease is the degree of generalized cerebral atrophy, revealed by MRI. The most important step in medical image diagnosis is image segmentation, which leads to precise extraction and classification. It relies on thresholding, based on the best threshold value detection that separates the background from the foreground. To estimate the best threshold value, Minimum Cross Entropy Thresholding is one of the well-known thresholding technique. To detect an optimal threshold that extracts the region reflecting the presence of the disease, we developed a novel segmentation algorithm for MRI images, using Pal method and based on a heterogeneous cross entropy thresholding technique. This technique consists of using a heterogeneous combination of 2 different statistical distributions to detect the optimal thresholding. On each image, we applied one type of distribution on the foreground and another on the background. We tried all the combinations between the statistical distributions Gaussian, Gamma, and Lognormal. Then, we confirmed validity of the proposed methodology, by using two benchmark of Alzheimer's disease images datasets, namely, OASIS-1 and OASIS-2. The hybrid combination methodology achieved more accuracy in detection, compared to other homogeneous segmentation techniques.