The prediction of brain health, specifically gray matter volume (GMV), is a critical area of research in understanding cognitive functions. Traditional methods to assess GMV are structural magnetic resonance imaging (MR), which requires expensive, time-consuming imaging techniques. With the availability of smartwatches and the possibility of tracking cardiorespiratory signals using these devices, we want to predict GMV using this data as an easier and less expensive method.
In this study, we’re aiming to explore whether cardiorespiratory signals, specifically photoplethysmography (PPG)-derived heart rate variability (HRV) features, can be used to predict GMV via machine learning (ML) models. So the main question of this research is “Can HRV-derived features accurately predict GMV?”.
Our key objectives are:
1. Create a database of HRV data from different subjects, alongside other personal information (like gender, age, weight, etc.) and GMV – This data will be a combination of UK BioBank data and a smaller dataset that has already been collected from individuals using smartwatch HRV data and GMV results from fMRI scans. Access to UK biobank data can greatly help us validate our results and improve the AI models used.
2. Use feature extraction methods to create a set of relevant and key features that will likely map the subject’s personalized data to GMV.
3. Use machine learning methods (ANN or RL) to train a model using these features.
4. Assess this method’s potential and accuracy as a more affordable screening tool for public health.
