Independent component analysis-based multimodal classification of Alzheimer´s disease versus healthy controls

Effective and accurate detection of Alzheimer´s disease (AD) at its early stage has been being paid more and more attention. However, most of existing researches have just focused on a single modality of imaging for diagnosis of AD. Although A single modality might be provides some meaningful information for diagnosing the disease, it still is difficult to meet the need of clinical diagnosis. With the fast development of multi-modality imaging such as structural magnetic resonance imaging (sMRI) and positron emission tomography (PET), combination of multi-modality has an opportunity to provide much more useful information to enhance the better performance of diagnosis of AD than that of a single one alone. The integration of the complementary information afforded by multimodal imaging protocols into a comprehensive analysis strategy is likely to aid in better discrimination and staging of AD. In the study, we proposed a method based on independent component analysis and support vector machine by combining sMRI and PET images to carry out the classification of AD subjects from healthy controls (HC). The experimental results illustrated that the classification between AD and HC subjects from the Alzheimer´s Disease Neuroimaging Initiative database was obtained with the averaged accuracy of 96.53% for Multimodal images at baseline, comparing to 88.95% for only sMRI images and 89.44% for only PET images. The multimodal classification approach hence allows more accurate early detection of brain abnormalities with greater sensitivity.