Evaluation and Prediction of Early Alzheimer’s Disease Using a Machine Learning-based Optimized Combination-Feature Set on Gray Matter Volume and Quantitative Susceptibility Mapping

Background: Because Alzheimer’s Disease (AD) has very complicated pattern changes, it is difficult to evaluate it with a specific factor. Recently, novel machine learning methods have been applied to solve limitations. Objective: The objective of this study was to investigate the approach of classification and prediction methods using the Machine Learning (ML)-based Optimized Combination-Feature (OCF) set on Gray Matter Volume (GMV) and Quantitative Susceptibility Mapping (QSM) in the subjects of Cognitive Normal (CN) elderly, Amnestic Mild Cognitive Impairment (aMCI), and mild and moderate AD. Materials and Methods: 57 subjects were included: 19 CN, 19 aMCI, and 19 AD with GMV and QSM. Regions-of-Interest (ROIs) were defined at the well-known regions for rich iron contents and amyloid accumulation areas in the AD brain. To differentiate the three subject groups, the Support Vector Machine (SVM) with the three different kernels and with the OCF set was conducted with GMV and QSM values. To predict the aMCI stage, regression-based ML models were performed with the OCF set. The result of prediction was compared with the accuracy of clinical data. Results: In the group classification between CN and aMCI, the highest accuracy was shown using the combination of GMVs (hippocampus and entorhinal cortex) and QSMs (hippocampus and pulvinar) data using the 2nd SVM classifier (AUC = 0.94). In the group classification between aMCI and AD, the highest accuracy was shown using the combination of GMVs (amygdala, entorhinal cortex, and posterior cingulate cortex) and QSMs (hippocampus and pulvinar) data using the 2nd SVM classifier (AUC = 0.93). In the group classification between CN and AD, the highest accuracy was shown using the combination of GMVs (amygdala, entorhinal cortex, and posterior cingulate cortex) and QSMs (hippocampus and pulvinar) data using the 2nd SVM classifier (AUC = 0.99). To predict aMCI from CN, the exponential Gaussian process regression model with the OCF set using GMV and QSM data was shown the most similar result (RMSE = 0.371) to clinical data (RMSE = 0.319). Conclusion: The proposed OCF based ML approach with GMV and QSM was shown the effective performance of the subject group classification and prediction for aMCI stage. Therefore, it can be used as personalized analysis or diagnostic aid program for diagnosis.