AUTHOR=Kang Li , Jiang Jingwan , Huang Jianjun , Zhang Tijiang TITLE=Identifying Early Mild Cognitive Impairment by Multi-Modality MRI-Based Deep Learning JOURNAL=Frontiers in Aging Neuroscience VOLUME=Volume 12 - 2020 YEAR=2020 URL=https://www.frontiersin.org/journals/aging-neuroscience/articles/10.3389/fnagi.2020.00206 DOI=10.3389/fnagi.2020.00206 ISSN=1663-4365 ABSTRACT=Mild cognitive impairment (MCI) is a clinical state having high risk of conversion to Alzheimers Disease(AD). Sincethere is no effective treatment to AD, it is extremely important to diagnose MCI as early as possible, which makes it possible to delay its progression towards AD. However, its challenging to identify early MCI(EMCI) because there are only mild changes of brain structure in the patients compared to normal control(NC). To extract remarkable features for these mild changes, in this paper, a multi-modality diagnosis approach based on deep learning was presented. Firstly, we propose to use structure MRI and diffusion tensor imaging(DTI) images as the multi-modality data to identify EMCI. Then, a convolutional neural network based on transfer learning technique is developed to extract features of the multi-modality data, where L1-norm is introduced to reduce the feature dimensionality and retrieve essential features for the identification. At last, the classifier produces 94.2% accuracy for EMCI versus NC on ADNI dataset. Experiment results show that multi-modality data can provide more useful information to distinguish EMCI from NC compared with single modality data and the proposed method can improve classification performance, which is beneficial to early intervention of AD. In addition, it is found that DTI image can act as an important biomarker for EMCI from the point of view of clinical diagnosis.