P4‐588: END‐TO‐END 3D‐CONVOLUTIONAL NEURAL NETWORK FOR PREDICTING CONVERSION FROM MILD COGNITIVE IMPAIRMENT TO ALZHEIMER'S DEMENTIA

• DOI: 10.1016/j.jalz.2019.08.136
• Publication Date: 2019-10-18T12:04:17Z
• Authors: Jinhyeong Bae; Jane Stocks; Ashley Heywood; Youngmoon Jung; Popuri Karteek; Mirza Faisal Beg; Lei Wang
• Keywords: 03 medical and health sciences; 0302 clinical medicine

Predicting conversion to Alzheimer's Dementia (AD) among Mild Cognitive Impairment (MCI) patients is invaluable for patient care, as well in selection clinical trials. This project utilized data from the Disease Neuroimaging Initiative (ADNI) develop end-to-end 3D-Convolutional Neural Network (3D-CNN) models classify subjects who did not progress AD (i.e., stable MCI or sMCI) vs. progressive pMCI). 470 sMCI and 293 pMCI patients’ skull-stripped baseline structural MRI (sMRI) scans were collected ADNI pre-processed by using crop, pad, bias field correction, affine linear alignment with FMRIB Software Library (FSL). Three separate 3D-CNN approaches developed compared. First, ImageNet implemented, Residual Network50 an augmentation Squeeze-Excitation Network, i.e. SE_ResNet50. Second, transfer learning was implemented. The domain knowledge of classifying Normal Control (sNC) Type (sDAT) transferred a reduced version ResNet50, i.e., ResNet35. Third, customized ResNet29 cResNet29, where residual block consisting modified drop ungeneralizable features. cResNet29 produced highest test classification accuracy, 72.81% SE_ResNet50 recorded second 70.01%. Based on these results, we confirmed that customizing ResNet performed better than augmenting additional Network. Transfer resulted 67.54%. It might imply sNC sDAT helpful pMCI.