Early Prediction of Alzheimer's Disease Using Artificial Intelligence and Cortical Features on T1WI Sequences

Rong Zeng ¹ Beisheng Yang ^2* Faqi Wu ^3* Huan Liu ^4* Xiaojia Wu ^2* Lin Tang ⁵ Rao Song ^2* Qingqing Zheng ^2* Xia Wang ^6* Dajing Guo ^2*

• ¹ Department of Radiology, Third Affiliated Hospital of Chongqing Medical University, Chongqing, China
• ² Department of Radiology, Second Affiliated Hospital of Chongqing Medical University, Chongqing, China
• ³ Department of Medical Service, Yanzhuang Central Hospital of Gangcheng District, Jinan, China, chongqing, China
• ⁴ GE Healthcare, Shanghai, Shanghai Municipality, China
• ⁵ Department of Radiology, Cancer Hospital, Chongqing University, Chongqing, China
• ⁶ Department of Radiology, Chongqing Western Hospital, Chongqing, China., chongqing, China

Accurately predicting the progression of mild cognitive impairment (MCI) to Alzheimer's disease (AD) is a challenging task, but it is crucial for helping to develop personalized treatment plans to improve prognosis. The aim of this study was to develop new technology for the early prediction of AD progression using artificial intelligence and cortical features on magnetic resonance imaging (MRI). A total of 162 MCI patients from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database were included. T1W images were acquired for each patient using a 3D-MPRAGE sequence. Patients were randomly divided into a training set (n = 112) and a validation set (n = 50) at a ratio of 7:3. Morphological features of the cerebral cortex were extracted with FreeSurfer software. Network features of gray matter were extracted with the GRETNA toolbox. Network, morphology, networkclinical, morphology-clinical, morphology-network and morphology-network-clinical models based on a multivariate Cox proportional hazards model were developed. The performance of each model was assessed with the concordance index (C-index). In the training group, the C-index values of the network, morphology, network-clinical, morphology-clinical, morphology-network and morphology-network-clinical models were 0.834, 0.926, 0.915, 0.949, 0.928, and 0.951, respectively; the respective values in the validation group were 0.765, 0.784, 0.849, 0.877, 0.884 and 0.880. The morphology-network-clinical model exhibited the best performance. A multi-predictor nomogram with high accuracy for individual AD prediction (C-index = 0.951) was established. Our resultssuggested that the early occurrence of AD could be accurately predicted using our morphologynetwork-clinical model and the multi-predictor nomogram. This model could help doctors make early and personalized treatment decisions in clinical practice, which has important clinical significance.