Radiomic Features of the Hippocampus for Diagnosing Early-Onset and Late-Onset Alzheimer's Disease

Yang Du; Shaowei Zhang; Yuan Fang; Qi Qiu; Lu Zhao; Wenjing Wei; Yingying Tang; Xia Li

doi:10.3389/fnagi.2021.789099

Radiomic Features of the Hippocampus for Diagnosing Early-Onset and Late-Onset Alzheimer's Disease

Front Aging Neurosci. 2022 Jan 26:13:789099. doi: 10.3389/fnagi.2021.789099. eCollection 2021.

Authors

Yang Du^{1

2}, Shaowei Zhang^{1

2}, Yuan Fang¹, Qi Qiu¹, Lu Zhao¹, Wenjing Wei¹, Yingying Tang³, Xia Li^{1

2}

Affiliations

¹ Department of Geriatric Psychiatry, Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
² Alzheimer's Disease and Related Disorders Center, Shanghai Jiao Tong University, Shanghai, China.
³ Shanghai Key Laboratory of Psychotic Disorders, Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine, Shanghai, China.

Abstract

Background: Late-onset Alzheimer's disease (LOAD) and early-onset Alzheimer's disease (EOAD) are different subtypes of AD. This study aimed to build and validate radiomics models of the hippocampus for EOAD and young controls (YCs), LOAD and old controls (OCs), as well as EOAD and LOAD. Methods: Thirty-six EOAD patients, 36 LOAD patients, 36 YCs, and 36 OCs from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database were enrolled and allocated to training and test sets of the EOAD-YC groups, LOAD-OC groups, and EOAD-LOAD groups. Independent external validation sets including 15 EOAD patients, 15 LOAD patients, 15 YCs, and 15 OCs from Shanghai Mental Health Center were constructed, respectively. Bilateral hippocampal segmentation and feature extraction were performed for each subject, and the least absolute shrinkage and selection operator (LASSO) method was used to select radiomic features. Support vector machine (SVM) models were constructed based on the identified features to distinguish EOAD from YC subjects, LOAD from OC subjects, and EOAD from LOAD subjects. The areas under the receiver operating characteristic curves (AUCs) were used to evaluate the performance of the models. Results: Three, three, and four features were selected for EOAD and YC subjects, LOAD and OC subjects, and EOAD and LOAD subjects, respectively. The AUC and accuracy of the SVM model were 0.90 and 0.77 in the test set and 0.91 and 0.87 in the validation set for EOAD and YC subjects, respectively; for LOAD and OC subjects, the AUC and accuracy were 0.94 and 0.86 in the test set and 0.92 and 0.78 in the validation set, respectively. For the SVM model of EOAD and LOAD subjects, the AUC was 0.87 and the accuracy was 0.79 in the test set; additionally, the AUC was 0.86 and the accuracy was 0.77 in the validation set. Conclusion: The findings of this study provide insights into the potential of hippocampal radiomic features as biomarkers to diagnose EOAD and LOAD. This study is the first to show that SVM classification analysis based on hippocampal radiomic features is a valuable method for clinical applications in EOAD.

Keywords: early-onset Alzheimer’s disease; hippocampus; late-onset Alzheimer’s disease; radiomics; support vector machine.