Title : Deep learning-based risk assessment of cognitive impairment using health examination data
Abstract:
For effective treatment and prevention of dementia, it is important to develop an objective, accurate, and inexpensive screening test for early diagnosis of cognitive impairment. To resolve these issues, we have developed a deep neural network (DNN) model that can estimate risk of cognitive impairment based on basic blood test data that do not contain dementia-specific biomarkers. The model was based on the relationship between cognitive function and systemic metabolic disorders in the elderly. That is, lifestyle-related diseases can result in vascular cognitive impairment (VCI) due to atherosclerosis. VCI plays an important role not only in vascular dementia, but also in the development of dementia in the elderly with Alzheimer's disease (AD). In addition, malnutrition, anemia, diabetes mellitus, liver dysfunction, and renal dysfunction can cause cognitive impairment and increase the risk of dementia. Importantly, these systemic metabolic disorders, including lifestyle-related disorders, can be detected by basic blood tests for health examinations that do not contain dementia-specific biomarkers. the estimation accuracy of the DNN model was validated in subjects who were not included in the training of the DNN model (r = 0.66, p < 0.001). The binary classification based on MMSE scores (cut-off value of 23/24) showed a high estimation accuracy (75%) and specificity (87%). In addition, we evaluated whether the DNN model makes it possible to estimate cerebral atrophy based on basic blood test data. Like the DNN model for estimating cognitive function, the input data used were subject age and basic blood test data. As the output of the DNN model, we used the MRI-based brain health quotient (BHQ), which measures the amount of gray matter (GM-BHQ) and the proportion of white matter anisotropy, which was used as an indicator of cerebral atrophy in this study. The DNN model was trained using brain docking data. We found a high estimation accuracy of the DNN model using repeated 5-Fold cross-validation. Moreover, even if only blood data were input without including age in the input data, the estimation accuracy was high. These results suggest that the DNN model allows us to estimate cognitive dysfunction and cerebral atrophy with high accuracy using basic blood test data, which does not include dementia-related biomarkers such as amyloid β.
Audience Takeaway:
- The audience will pay more attention to the systemic metabolic status of dementia patients by understanding the close relationship between systemic metabolic disorders and cognitive function and brain atrophy.
- In order to reduce risk factor for dementia onset, the audience will put more effort into improving systemic metabolic disorders such as lifestyle-related diseases.
- I believe that the application of deep learning to clinical dementia will progress.
- This research will become a new mass screening method for dementia and contribute to the early detection and prevention of dementia.
- It will be possible to select high-risk dementia patients more accurately and efficiently than the interview-type tests (such as MMSE) currently used for dementia screening.
- Other benefits include 1) it uses only data of health checkup including blood data, thus, no need of blood sampling, 2) it does not use a special equipment for measurering dementia-specific biomarkers, thus, not expensive, 3) it allows to use smartphone for personal assessment dementia risk personally, 4) Individual risk of dementia can be known from blood data findings, making it possible to provide personalized dietary guidance.
