Robust Deep Learning for Improved Classification of AD/MCI Patients Feng Li 1 , Loc Tran 1 , Kim-Han Thung 2 , Shuiwang Ji 3 , Dinggang Shen 2 , and Jiang Li 1 1 Department of ECE, Old Dominion University, Norfolk, VA 2 Department of Radiology, University of North Carolina at Chapel Hill, NC 3 Department of Computer Science, Old Dominion University, Norfolk, VA Abstract. Accurate classification of Alzheimer’s Disease (AD) and its prodromal stage, Mild Cognitive Impairment (MCI), plays a critical role in preventing progression of memory impairment and improving quality of life for AD patients. Among many research tasks, it is of particular interest to identify noninvasive imaging biomarkers for AD diagnosis. In this paper, we present a robust deep learning system to identify differ- ent progression stages of AD patients based on MRI and PET scans. We utilized the dropout technique to improve classical deep learning by preventing its weight co-adaptation, which is a typical cause of over- fitting in deep learning. In addition, we incorporated stability selection, an adaptive learning factor and a multi-task learning strategy into the deep learning framework. We applied the proposed method to the ADNI data set and conducted experiments for AD and MCI conversion diag- nosis. Experimental results showed that the dropout technique is very effective in AD diagnosis, improving the classification accuracies by 6.2% on average as compared to classical deep learning methods. 1 Introduction Alzheimer’s disease is the sixth-leading cause of death in the United States [1]. AD patients usually undergo progressive stages of cognitive and memory func- tion impairment, including prodromal, MCI and AD. For each of these stages, significant amount of research has been conducted aiming to understanding the underlying pathological mechanisms. In addition, imaging biomarkers have been identified using different imaging modalities such as magnetic resonance imaging (MRI) [2], positron emission tomography (PET) [3], and functional MRI (fMRI) [4]. Imaging biomarkers are a set of indicators computed from image modalities and can be used for early detection of AD disease. It has been shown that fusing these different modalities may lead to more effective imaging biomarkers [5]. Deep learning is a new breakthrough in machine learning. The first successful deep learning framework, auto-encoder, was developed in 2006 [6]. It was subse- quently used in other application fields and achieved state-of-the-art performance in speech recognition, image classification and computer vision [7]. Deep learning itself also evolves after 2006. For instance, the multimodal deep learning frame- work boosted speech classification by learning a shared representation between G. Wu et al. (Eds.): MLMI 2014, LNCS 8679, pp. 240–247, 2014. c  Springer International Publishing Switzerland 2014