Early diagnosis of Alzheimer’s disease based on Partial Least Squares and Support Vector Machine

An accurate and early diagnosis of the Alzheimer’s disease (AD) is of fundamental importance for the patient medical treatment. It has been shown that pathological manifestations of AD may be detected thought functional images even before that the patients becomes symptomatic. This fact has led researchers to propose new ways for analyzing functional data in order to get more accurate Computer-Aided Diagnosis (CAD) systems for this disorder. In this paper we show an effective approach for Single Photon Emission Computed Tomography feature extraction that improves the accuracy of CAD systems for AD. The proposed methodology uses a Partial Least Squares algorithm for extracting score vectors and the Out-Of-Bag error for selecting a number of scores that are used as features. Subsequently, a Support Vector Machine (SVM) based classifier determines the underlying class of the images, thus performing diagnostics. In order to test this approach we have used an image database for AD with 97 SPECT images from controls and AD patients. The images were visually labeled by experienced clinicians after the properly normalization. Several experiments have been developed to compare the proposed methodology and previous approaches. The results show that our method yields accuracy rates over 90%, outperforming several recently reported CAD systems for AD diagnosis.