![[2509.12253] Physics-Informed Neural Networks vs. Physics Models for Non-Invasive Glucose Monitoring: A Comparative Study Under Noise-Stressed Synthetic Conditions](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2509.12253] Physics-Informed Neural Networks vs. Physics Models for Non-Invasive Glucose Monitoring: A Comparative Study Under Noise-Stressed Synthetic Conditions
Summary
This study compares Physics-Informed Neural Networks (PINNs) and traditional physics models for non-invasive glucose monitoring under noise-stressed conditions, revealing that physics-engineered models can outperform complex neural networks in low signal-to-noise scenarios.
Why It Matters
Non-invasive glucose monitoring is crucial for diabetes management, yet current methods struggle with noise interference. This research highlights the potential of physics-informed approaches to enhance accuracy in challenging conditions, which could lead to better health outcomes for patients.
Key Takeaways
- Physics-engineered models can achieve lower error rates than complex neural networks in low-SNR conditions.
- The study introduces a noise-stressed NIR simulator to benchmark various glucose monitoring methods.
- Enhanced Beer-Lambert model outperformed PINNs with fewer parameters and faster inference times.
- Carefully engineered physics features are essential for effective noise suppression.
- The findings suggest a paradigm shift in how glucose monitoring challenges are approached.
Electrical Engineering and Systems Science > Image and Video Processing arXiv:2509.12253 (eess) [Submitted on 12 Sep 2025 (v1), last revised 20 Feb 2026 (this version, v2)] Title:Physics-Informed Neural Networks vs. Physics Models for Non-Invasive Glucose Monitoring: A Comparative Study Under Noise-Stressed Synthetic Conditions Authors:Riyaadh Gani View a PDF of the paper titled Physics-Informed Neural Networks vs. Physics Models for Non-Invasive Glucose Monitoring: A Comparative Study Under Noise-Stressed Synthetic Conditions, by Riyaadh Gani View PDF HTML (experimental) Abstract:Non-invasive glucose monitoring outside controlled settings is dominated by low signal-to-noise ratio (SNR): hardware drift, environmental variation, and physiology suppress the glucose signature in NIR signals. We present a noise-stressed NIR simulator that injects 12-bit ADC quantisation, LED drift, photodiode dark noise, temperature/humidity variation, contact-pressure noise, Fitzpatrick I-VI melanin, and glucose variability to create a low-correlation regime (rho_glucose-NIR = 0.21). Using this platform, we benchmark six methods: Enhanced Beer-Lambert (physics-engineered ridge regression), Original PINN, Optimised PINN, RTE-inspired PINN, Selective RTE PINN, and a shallow DNN. The physics-engineered Beer Lambert model achieves the lowest error (13.6 mg/dL RMSE) with only 56 parameters and 0.01 ms inference, outperforming deeper PINNs and the SDNN baseline under low-SNR conditions. The study r...
