A pure software, host agnostic signal processing core developed to eliminate systemic skin pigmentation inaccuracies in dual wavelength pulse oximetry systems.
Standard commercial pulse oximeters measure blood oxygen saturation (SpO2) by analysing how oxygenated and deoxygenated haemoglobin absorb red (660nm) and infrared (940nm) light differently. However, melanin, the pigment determining skin tone, absorbs red light significantly more than infrared light, distorting the optical ratio and causing systematic SpO2 overestimation in darker skinned patients across Fitzpatrick Types IV-VI and the Monk Skin Tone scale.
This causes occult hypoxemia, a clinically dangerous condition where a monitor falsely reads safe blood oxygen levels (≥92%) while true arterial saturation is critically low, leading to delayed treatment decisions, documented in FDA safety communications since 2022.
Rather than requiring multi wavelength hardware redesigns, a pure mathematical compensation engine was developed to optimize existing hardware. By evaluating realtime non-pulsatile optical attenuation, the algorithm dynamically classifies tissue properties and balances optical ratios without modifying any optoelectronic components.
The algorithm was verified using the PTT-PPG dual-wavelength dataset (PhysioNet), recorded with a MAX30101 medical-grade optical sensor at 500Hz.
| Metric | Value | Standard | Status |
|---|---|---|---|
| ARMS | 2.35% | FDA ≤3.0% | ✅ PASS |
| Precision (σ) | 0.22% | — | Excellent |
| Valid Windows | n=213 | — | Statistically sound |
| PI Quality Gate | 88% valid | — | Working correctly |
Skin aware correction eliminates dark skin tone bias without introducing measurement inaccuracy in light or medium skin tones (Fitzpatrick I-VI).
| Component | Detail |
|---|---|
| Design Environment | MATLAB R2023b / Simulink |
| Design Methodology | Model-Based Design (V-Model) |
| Code Output | Production C++ via Embedded Coder |
| Target Architecture | Host-agnostic, inherited sample-time (-1) |
| Regulatory Framework | IEC 62304 Class B SDLC |
| Dataset | PTT-PPG (PhysioNet), MAX30101 sensor |
✅ Requirements specification complete (URS + SRS, 50 requirements)
✅ Full bidirectional traceability (URS SRS Design Verification)
✅ MATLAB algorithm prototyping complete (6 blocks)
✅ Simulink Model-Based Design complete
✅ Production C++ code generation complete (Embedded Coder)
✅ Performance verification complete (ARMS = 2.35%)
✅ Full DHF documentation complete
This public repository contains verification visuals and high level architectural documentation only.
| File | Description |
|---|---|
| Full_Model.png | Simulink functional module architecture |
| Complete_Algorithm_Results_PTT_PPG.png | Validation metrics |
| DUAL-CHANNEL_AC-DC_DECOMPOSITION_PTT_PPG.png | Signal decomposition |
| Melanin_Index_Analysis_PTT_PPG.png | Tier classification metrics |
IP Notice: Proprietary model logic (
.slx), calibration parameter scripts (.m), correction coefficients, and generated C++ source files are explicitly excluded to safeguard intellectual property.
For technical discussions or collaboration enquiries:
Ifeanyichukwu Obidike