The Impact of Different Skin Colors on the Accuracy of Photoplethysmography (PPG) Oxygen Saturation Measurement Techniques and Research on Correction Algorithm

Photoplethysmography (PPG) is a non-invasive blood oxygen measurement technique widely used in medical monitoring and health assessment. This technology relies on the Lambert-Beer law to monitor blood oxygen saturation (SpO2) by analyzing the absorption changes of blood to specific wavelengths of light. In the classic blood oxygen saturation calculation model, SpO2 = A + BR, where A and B are fixed coefficients obtained after model fitting, and R = (ACir/DCir) / (ACred/DCred). ACir and ACred represent the alternating components of red and infrared light (the parts that change with pulse), while DCir and DCred represent the direct current components of these two types of light (the parts that do not change with pulse). Although PPG is favored in health tracking, sports science, and clinical diagnosis due to its convenience and non-invasive nature, this method is based on modeling under ideal conditions. In reality, biological tissue is a complex optical medium with strong scattering, weak absorption, and anisotropy, which does not fully comply with the Lambert-Beer law. In particular, the absorption and scattering effects of different skin pigments directly affect the accuracy of blood oxygen saturation measurement based on PPG, especially in low oxygen saturation conditions, and this deviation is particularly evident in individuals with darker skin tones. To address this issue, this study designed a hardware system, including two red and infrared light LEDs of different wavelengths and a photodiode (PD) sensor. The PD sensor simultaneously receives signals reflected from the two LED light sources, controls the two LEDs of different wavelengths through frequency division driving, and uses a fast phase-Locking algorithm to demodulate the PD sensor’s signal, achieving synchronous measurement of the two LEDs of different wavelengths. In addition, this study constructed an ex vivo vascular model, simulating human blood by injecting fat emulsion solution into the vascular model, and using a direct current pump to act as a “heart” function, controlling the back-and-forth flow of the fat emulsion solution from the reservoir to the simulated blood vessels, simulating human blood flow. By pasting neutral density filters and specific wavelength filters of different attenuation degrees on the surface of the simulated blood vessels, the absorption and scattering effects of different human skin colors on light were simulated. Finally, this study re-established a new blood oxygen saturation calculation model through a multiple linear regression model to improve the accuracy of PPG-based blood oxygen saturation measurement under different skin colors. In the experiment, the attenuation degree of the filters was adjusted to simulate 10 different levels of skin color depth, and the density of the fat emulsion solution was changed to simulate 10 different levels of blood oxygen saturation. In this way, a total of 100 sets of data were collected, each set containing paired information of corresponding skin color level and blood oxygen value. At the same time, this study also collected 10 sets of data of different levels of blood oxygen saturation under the condition of no filter as a control group. The study found that as the attenuation degree of the filter increased and the density of the fat emulsion solution decreased (simulating darker skin color and lower blood oxygen value), the calculation error of blood oxygen saturation also increased. Therefore, this study introduced two correction coefficients C and D into the classic blood oxygen value calculation model SpO2 = A + BR, resulting in a new calculation model: SpO2 = (A + BR + C × Attn) × (1 + D), where Attn represents the attenuation level of the filter. The values of A, B, C, and D were determined through a multiple linear regression model, while the R value remained consistent with the classic model, and the R value was calculated in real-time through the alternating and direct current components of red and infrared light collected by the hardware device. The results of the study showed that the error between the corrected blood oxygen measurement value and the actual simulated blood oxygen value was significantly reduced, with the maximum error reduced from 10% to 2%. This finding confirms the effectiveness of the correction algorithm in reducing the impact of skin color on blood oxygen measurement. By using an appropriate correction algorithm, the accuracy and reliability of blood oxygen measurement technology under different skin color differences can be effectively improved, which is of great significance for optimizing the design and application of blood oxygen measurement devices.