Clinical devices play a vital role in diagnosing and monitoring people’s health. A pulse oximeter (PO) is one of the most common clinical devices for critical medical care. In this paper, we explain how we developed a wearable PO. We propose a new electronic circuit based on an analog filter that can separate red and green photoplethysmography (PPG) signals, acquire clean PPG signals, and estimate the pulse rate (PR) and peripheral capillary oxygen saturation (SpO₂). We propose a PR and SpO₂ measurement algorithm with and without the motion artifact. We consider three types of motion artifacts with our acquired clean PPG signal from our proposed electronic circuit. To evaluate our proposed algorithm, we measured the accuracy of our estimated SpO₂ and PR. To evaluate the quality of our estimated PR (bpm) and SpO₂ (%) with and without the finger motion artifact, we used the quality evaluation metrics: mean absolute percentage error (MAPE), mean absolute error (MAE), and reference closeness factor (RCF). Without the finger motion condition, we found that our proposed wearable PO device achieved an average 2.81% MAPE, 2.08 bpm MAE, 0.97 RCF, and 98.96% SpO₂ accuracy. With a finger motion, the proposed wearable PO device achieved an average 4.5% MAPE, 3.66 bpm MAE, 0.96 RCF, and 96.88% SpO₂ accuracy. We also show a comparison of our proposed PO device with a commercial Fingertip PO (FPO) device. We have found that our proposed PO device performs better than the commercial FPO device under finger motion conditions. To demonstrate the implementation of our wearable PO, we developed a smartphone app to allow the PO device to share PPG signals, PR, and SpO₂ through Bluetooth communication. We also show the possible applications of our proposed PO as a wearable, hand-held PO device, and a PPG signal acquisition system.

Glycated hemoglobin and blood oxygenation are the two most important factors for monitoring a patient’s average blood glucose and blood oxygen levels. Digital volume pulse acquisition is a convenient method, even for a person with no previous training or experience, can be utilized to estimate the two abovementioned physiological parameters. The physiological basis assumptions are utilized to develop two-finger models for estimating the percent glycated hemoglobin and blood oxygenation levels. The first model consists of a blood-vessel-only hypothesis, whereas the second model is based on a whole-finger model system. The two gray-box systems were validated on diabetic and nondiabetic patients. The mean absolute errors for the percent glycated hemoglobin (%HbA1c) and percent oxygen saturation (%SpO₂) were 0.375 and 1.676 for the blood-vessel model and 0.271 and 1.395 for the whole-finger model, respectively. The repeatability analysis indicated that these models resulted in a mean percent coefficient of variation (%CV) of 2.08% and 1.74% for %HbA1c and 0.54% and 0.49% for %SpO₂ in the respective models. Herein, both models exhibited similar performances (HbA1c estimation Pearson’s R values were 0.92 and 0.96, respectively), despite the model assumptions differing greatly. The bias values in the Bland–Altman analysis for both models were – 0.03 ± 0.458 and – 0.063 ± 0.326 for HbA1c estimation, and 0.178 ± 2.002 and – 0.246 ± 1.69 for SpO2 estimation, respectively. Both models have a very high potential for use in real-world scenarios. The whole-finger model with a lower standard deviation in bias and higher Pearson’s R value performs better in terms of higher precision and accuracy than the blood-vessel model.

Effective management of diabetes requires accurate monitoring of blood glucose levels. Traditional invasive methods for such monitoring can be cumbersome and uncomfortable for patients. In this study, we introduce a noninvasive approach to estimate blood glucose levels using photoplethysmography (PPG) signals. We have focused on blood glucose prediction using wrist PPG signals and explored various PPG waveform-based features, including AC to DC ratio (AC/DC) and intrinsic mode function (IMF)-based features derived from empirical mode decomposition (EMD). To the best of our knowledge, no studies have been found using EMD-based features to estimate blood glucose levels noninvasively. Additionally, feature importance-based selection has also been used to further improve the accuracy of the proposed model. Among the four machine learning algorithms considered in this study, CatBoost consistently outperformed XGBoost, LightGBM, and random forest across a wide number of features. The best performing model, CatBoost, achieved Pearson’s r of 0.96, MSE 0.08, ${\mathrm{R}}^2$ score 0.92, and MAE 8.01 when considering the top 50 features selected from both PPG waveform-based features and IMF-based features. The p-values for all models were <0.001, indicating statistically significant correlations. Overall, this study provides valuable insights into the feasibility and effectiveness of noninvasive blood glucose monitoring using advanced machine learning techniques.