A hierarchical model for identifying mild cognitive impairment

As very common disorder, Alzheimer´s disease (AD) brings many health problems for the older people. Predicting mild cognitive impairment (MCI), a progression stage intermediated between normal status and dementia, in patients who have some symptoms of cognitive decline which can have significance influence on treatment choice. In this work, a novel hierarchical PCA-SVM (Principal component analysis-support vector machines) model to identify mild cognitive impairment using the information of Magnetic Resonance Imaging (MRI), Position Emission Tomography (PET)and the cerebrospinal fluid (CSF). In the first, 151 samples, including 99 MCI patients and 52 healthy controls, are represented by the MRI, PET, and CSF. An effective feature extraction method, PCA, which focuses on dimension reduction was then adopted to construct a new discriminative feature set. Finally, a predictor based on SVM method were trained and then used for identifying MCI participants from the normal people. Here the number of components will be determined by the prediction accuracy. Based on an optimal selection of component number, our proposed method achieved 70.5% prediction accuracy. To compare our method with state-of-the-art MCI identification methods, i.e., SVM and SVM with fusion kernels based algorithms, other comparison experiments are performed, and the experimental results demonstrate that our proposed PCA-based method is a powerful tools to predict MCI with excellent performance and high efficiency.