Unpacking PPG Signals for Transparent Blood Glucose Level Sensing.

Diabetes affects millions of people worldwide each year, posing a major challenge to global public health systems. The consequences of diabetes transcend the individual level, manifesting across familial structures and exerting a wide-ranging impact on society at large. Maintaining blood glucose levels within the normal range is critical; chronic dysregulation can lead to severe complications affecting vital organs and bodily functions. While lifestyle interventions such as diet regulation and physical activity play key roles in diabetes management, monitoring blood glucose regularly has proven to be helpful in maintaining it.

Among emerging approaches, photoplethysmography (PPG) has shown promise for estimating blood glucose levels (BGL) using machine learning (ML) techniques. PPG’s prevalence in wearable technology makes it a strong candidate for continuous, non-invasive glucose monitoring, especially in edge-computing scenarios where computational resources are limited. However, most of the existing work uses complex black-box approaches. In this thesis, you will work on identifying key PPG features that contribute to blood glucose level estimation.

Literature Review

• Summarize key studies on PPG-based blood glucose estimation.
• Identify studies that used interpretable models and/or focused on key factors that changed in PPG with changes in blood glucose level.
• Identify 5 published datasets that include (a) both PPG and glucose measurements and (b) multiple measurements across each participant.

Data Analysis

• Implement an extendable pipeline for data preprocessing and feature extraction for the identified datasets.
• Extract relevant features and find top features that correlate within- and between-subject cases.
• Identify key features that differ between diabetic and non-diabetic subjects.
• Build an interpretable model that evaluates PPG-based BGL level monitoring.

• Background knowledge in signal processing and biomedical sensing would be great
• Proficiency with Python for signal analysis and model development
• Familiarity with Statistical Tests and basic ML knowledge.
• Capability to document results clearly and interpret findings in the context of physiological sensing.