Deep Learning Framework for Alzheimer’s Disease Diagnosis via 3D-CNN and FSBi-LSTM

IEEE Access

Xiaohua Xiao¹  Huoyou Hu²  Feng Zhou³  Peng Yang⁴  Tianfu Wang⁴  Ahmed Elazab⁴  Chiyu Feng⁴  Baiying Lei⁵ 

[1] Affiliated Hospital of Shenzhen University, Health Science Center, Shenzhen University, Shenzhen Second People&x2019;Dearborn, Dearborn, MI, USA;Industrial and Manufacturing Systems Engineering, University of Michigan&x2013;National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering, Health Science Center, Shenzhen University, Shenzhen, China;s Hospital, Shenzhen, China;

关键词: Alzheimer’s disease;    3D-CNN;    FSBi-LSTM;    multi-modal fusion;   

DOI  :  10.1109/ACCESS.2019.2913847

来源: DOAJ

【 摘 要 】

Alzheimer's disease (AD) is an irreversible progressive neurodegenerative disorder. Mild cognitive impairment (MCI) is the prodromal state of AD, which is further classified into a progressive state (i.e., pMCI) and a stable state (i.e., sMCI). With the development of deep learning, the convolutional neural networks (CNNs) have made great progress in image recognition using magnetic resonance imaging (MRI) and positron emission tomography (PET) for AD diagnosis. However, due to the limited availability of these imaging data, it is still challenging to effectively use CNNs for AD diagnosis. Toward this end, we design a novel deep learning framework. Specifically, the virtues of 3D-CNN and fully stacked bidirectional long short-term memory (FSBi-LSTM) are exploited in our framework. First, we design a 3D-CNN architecture to derive deep feature representation from both MRI and PET. Then, we apply FSBi-LSTM on the hidden spatial information from deep feature maps to further improve its performance. Finally, we validate our method on the AD neuroimaging initiative (ADNI) dataset. Our method achieves average accuracies of 94.82%, 86.36%, and 65.35% for differentiating AD from normal control (NC), pMCI from NC, and sMCI from NC, respectively, and outperforms the related algorithms in the literature.

【 授权许可】

Unknown