Problem

Research has shown that pulse oximetry can produce falsely elevated oxygen saturation levels in individuals with higher melanin concentrations, leading to potential underdiagnosis of hypoxemia. This bias contributes to documented disparities in health outcomes—such as higher COVID-19 mortality rates among people of color compared to age-adjusted white counterparts.

To address these inaccuracies, our team aimed to design an electromechanical signal chain that could reliably convert photodiode current into a stable voltage while minimizing the influence of skin pigmentation, ambient light, and anatomical variation. Using the OP950 photodiode sensor, we developed a transimpedance amplifier (TIA) followed by an inverting low-pass and a non-inverting high-pass filter. Our goal was to achieve a passband sensitive to physiological signals while suppressing 60 Hz electromagnetic interference and 120 Hz ambient light fluctuations. We further documented signal processing workflows in Arduino IDE and C++ to benchmark our prototype against commercial gold-standard devices, with particular emphasis on improving accuracy and equity in SpO₂ measurement across skin tones.

Solution

We redesigned the signal path to replace the initial AD620-based TIA with a fully discrete op-amp stack, enabling precise current-to-voltage conversion and better bandwidth control. The transfer functions were tuned so that the overall system bandwidth approximated 0.48-19.4 Hz, aligning with Nyquist sampling theorem for capturing pulse dynamics while suppressing high-frequency noise. We validated stage-by-stage performance with oscilloscopes and function generators to ensure each block met expected specifications.

We then implemented signal processing documentation in Arduino IDE and Python to translate raw sensor data into SpO₂ estimates, incorporating skin tone adjustments via a Monk Skin Tone-based grayscale metric. By comparing AC/DC components of red and infrared signals and accounting for melanin absorption, we aimed to improve accuracy across diverse skin tones. The final design achieved a gain distribution that balanced signal integrity with the Arduino’s input range, providing a practical, in-lab demonstration of an in-house solution

Team

Eliza Coleto
- Led the development of the spectral processing and compensation algorithms for skin tone–dependent SpO₂ adjustments
- Designed and implemented Python and Arduino scripts for signal acquisition, filtering, and ratio-of-ratios SpO₂ computation
- Converted RGB skin tone images into grayscale and mapped them to the Monk Skin Tone scale to quantify melanin absorption
Jasmine Escobedo
- Redesigned the TIA and filter stages for robust current-to-voltage conversion
- Calculated and verified cutoff frequencies (0.48–19.4 Hz) and gain distribution across amplifier stages to optimize the signal chain
- Contributed to testing setup, including signal injection via function generator and gain measurement analysis
Josh Reyes
- Initiated the first iteration of the pulse oximeter circuit design and conducted lab validation of signal performance
- Supported circuit debugging and gain-stage testing during prototype evaluation
- Coordinated team milestones and deadlines to ensure timely progress across hardware development, data processing, and documentation

Left to right: Jasmine Escobedo, Josh Reyes, Eliza Coleto

Here is a picture of Jasmine and I with our uGSI, Lucile Zhu.

Context: Jasmine and I were Co-Vice Presidents for the Biomedical Engineering Society at Berkeley during our senior year, and Lucile was one of the Vice Presidents before us, so here is a lineage picture! Shoutout to Lucile for dealing with all of my crash outs for this class (I was messaging and emailing her too many times) 😭😭

Left to right: Josh Reyes, Lucile Zhu (top), Jasmine Escobedo

Slide Deck Presentation

https://docs.google.com/presentation/d/1pFJgX4r4Rd7Iid4xicWFlN7LLgOcpMAv95qORjXzuhQ/edit?usp=sharing

Project Report

https://docs.google.com/document/d/1SsZBQmdJ-BxdGo0-WlkW-BJlFWQ1MqoiCtOF13B1uks/edit?usp=sharing

Enhancing Pulse Oximeter & SpO2 Accuracy based on Skin Pigmentation