Radiomics features as predictors to distinguish fast and slow progression of Mild Cognitive Impairment to Alzheimer's disease

Prediction of Alzheimer's disease (AD) from Mild Cognitive Impairment (MCI) by analyzing Magnetic Resonance Imaging (MRI) image features has become popular in recent years. However, defining effective predictive biomarkers is still challengeable. The 'radiomics' is an established method to identify advanced and high order quantitative imaging features for computer-aided diagnosis and has been applied into oncology study. However, it has not been applied into brain disorder disease study. Therefore, the purpose of this study is to identify whether the features from radiomics could be the predictors of the conversion from MCI to AD. We analyzed 197 samples with MRI scans from the ADNI database, which contained 32 healthy subjects and 165 MCI patients. Firstly, we extracted 215 radiomics features from hippocampus. Then we used Cronbach's alpha coefficient, the intra-class correlation coefficient, Kaplan-Meier model and cox regression to select 44 radiomics features as effective features. Finally, we used SVM classification to validate these features. The results showed that the classification accuracy using linear, polynomial and sigmoid kernel could achieve 80.0%, 93.3% and 86.6% to distinguish MCI-to-AD fast and slow converter. As a result, this study indicated that the radiomics features are potential to be applied into predicting AD from MCI.