P4-287: prediction of 7-year progression from subjective cognitive decline to mci: evidence from the china longitudinal ageing study (clas)

It is increasingly important to identify which subjective cognitive decline (SCD) would eventually develop cognitive impairment many years later. However, long-term structural MRI study investigating the conversion from SCD to MCI is scarce. The present work included 76 SCD individuals with two neuropsychological evaluations (baseline and 7 years follow-up) and a T1-MRI scan at baseline. All the data were from the China Longitudinal Aging Study (CLAS) started on 2011. We proposed a classification model to discriminate progressive SCD (pSCD) who had converted to MCI (n=24) at the follow-up from stable SCD (sSCD) (n=52). Multi-type features including clinical information, physical disease, lifestyle, psychological measurement and FreeSurfer-derived MRI features were fed into our prediction model. To fully consider the interaction between the features, Relief algorithm and cost sensitive support vector machine (SVM) classifiers are wrapped for feature selection and classification (Fig. 1). Of total 223 features, 5 features were selected as the most contributive features in our study. They are: stroke history, years of education, baseline MoCA score, volume of left amygdala, and right white matter in banks of superior temporal sulcus (wmbankssts) (Fig. 2). Their combination achieves a satisfactory prediction with an Area Under the Curve (AUC) score of 0.805 (Fig. 3). More interestingly, decreased gray matter volume in left amygdala and increased right wmbankssts were both found in pSCD compared to sSCD. Further group analysis showed larger wmbankssts in the two progressive groups compared to the stable groups, which may indicate a compensatory hypothesis (Fig. 4). Additional independent DTI study using 15 healthy elderly community individuals data proved that neurons in bankssts are linked to all most every brain lobe throughout tracts of axons, including hippocampus and amygdala (Fig. 5).