NeuroAI: AI‐based prediction model of Amyloid PET positivity using MR images and APOE genotype

Abstract

AbstractBackgroundSince acquiring amyloid PET data is costly and long waiting period due to the inaccessibility of PET machine, predicting positivity of amyloid PET data has many benefits. Therefore, we aimed to develop an artificial intelligence (AI) model for predicting amyoloid PET positivity for mild cognitive impairment (MCI) subjects with MR images and APOE genotype.MethodA total 1,742 MCI subjects were enrolled in the Gwangju Alzheimer’s and Related Dementia cohort registry, Korea. We trained the deep learning model using MR images (T1 and Flair) and their APOE genotyping data. Finally, we conducted an external validation with 227 MCI subjects (118 with Aβ(+) and 109 with Aβ(‐)) from Chonnam National University Hospital and Chosun University Hospital in Korea.ResultWith independent test set, our model had the great performance with an AUC of 0.84 (0.78‐0.89), Accuracy of 78% (72‐83%), Sensitivity of 73% (64‐81%), and Specificity of 83% (74‐89%). The Korea Ministry of Food and Drug Safety granted clearance for the Phase III clinical trail of our model (NeuroAI) in March 2023 (NCT05383053).ConclusionOur model could help reduce the need for costly and time‐consuming amyloid PET scans, while also improving the accuracy of Alzheimer’s Disease diagnosis.