Brain Volumetry‐based Amyloid PET Positivity Prediction in Subjective Cognitive Decline Patients
Wooseok Jung, Dong Won Yang, Chunghwee Lee, Yun Jeong Hong, SeongHee Ho, Jee Hyang Jeong, Kee Hyung Park, SangYun Kim, Min Jeong Wang, Seong Hye Choi, Dongsoo Lee, Junghyun Kang, Jinyoung Kim, Yeha Lee- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
Amyloid beta PET screening for subjective cognitive decline (SCD) patients would identify a risk factor of Alzheimer’s disease (AD) at an early stage but is generally infeasible due to the high cost and radiation exposure. Brain volumetry using 3D T1 MRI integrated with genetic information can alternatively suggest amyloid positivity in SCD subjects.
Method
Individuals of age between 60 and 85 feeling cognitive decline and scoring 7% to 50% on the memory test while exceeding 7% of the other tests comprises the SCD dataset via Seoul Neuropsychological Battery (SNSB) scheme (n = 119; 26 Aß+; 24 APOE4+). We measured the volume, volume to intracranial volume ratio, cortical thickness and their normative percentiles of 104 brain subregions, respectively, using VUNO‐Med DeepBrain, an automated brain volumetry and neurodegenerative disease diagnosis‐supporting software. The significance level is set to a = 0.005 to find differentiating regions between amyloid positive and negative groups. Furthermore, four machine learning models (Logistic Regression, Random Forest, XGBoost, TabNet) were trained using the combination of APOE4 information, age, and each of the six different volumetry results to predict amyloid positivity in SCD dataset. The training dataset (n = 102; 36 Aß+; 31 APOE4+) consists of subjects from significant memory concern (SMC) cohort of ADNI2 and ADNIGO studies.
Result
Swell of the choroid plexus and shrink of the lingual cortex (both p < .005) are statistically significant in Aß+ SCD patients compared to the Ab‐ patients. The random forest model trained by the volume data plus APOE4 and age information showed the highest performance (accuracy = 0.79, AUROC = 0.75, precision = 0.81, recall = 0.79, F1 = 0.80) in the amyloid positivity prediction task among the 24 models, higher than the model without APOE4 information (accuracy = 0.70, AUROC = 0.71, precision = 0.77, recall = 0.70, F1 = 0.72).
Conclusion
The result indicates a statistically significant difference in choroid plexus volume and lingual cortex thickness between Aß+ and Aß‐ SCD patients. In addition, using both APOE e4 information and brain volumetry is crucial for amyloid positivity prediction. This study was supported by a grant from the Ministry of Health and Welfare, HI18C0530.