Key result
Fingertip PPG-based Random Forest model estimates fasting blood glucose within ~/dL of reference values.
Why the study?
Routine monitoring of diabetes mellitus and hypertension relies on fingerstick glucometry and pneumatic cuffs, which limit adherence. This study evaluated a low-cost fingertip photoplethysmogram prototype for non-invasive estimation of heart rate, peripheral oxygen saturation, blood pressure, and blood glucose.
Does a low-cost single-fingertip photoplethysmogram prototype accurately estimate heart rate, SpO2, blood pressure, and blood glucose compared to standard reference devices in fasting adult volunteers?
Observational (n=30)
No
Does a low-cost single-fingertip photoplethysmogram prototype accurately estimate heart rate, SpO2, blood pressure, and blood glucose compared to standard reference devices in fasting adult volunteers?
This preliminary engineering prototype demonstrates the technical feasibility of extracting heart rate, SpO2, blood pressure, and blood glucose features from a single fingertip PPG sensor, though small sample size and methodological limitations preclude any clinical application without rigorous subject-independent validation.
Answers are grounded in this paper and cite their sources.
Does not support clinical glucose estimation from fingertip PPG; leaves open technical feasibility for multi-parameter sensors pending validation.
Diabetes mellitus and hypertension are leading causes of morbidity and premature mortality, yet routine monitoring depends on fingerstick glucometry and pneumatic cuffs, which limit adherence. This paper presents the design and a preliminary, small-cohort evaluation of a low-cost prototype that derives heart rate (HR), peripheral oxygen saturation (SpO2), systolic/diastolic blood pressure (SBP/DBP), and blood glucose (BG) estimates from a single fingertip photoplethysmogram (PPG). The hardware combines a MAX30102 dual-wavelength sensor (660 nm red, 940 nm infrared), an ATmega328P-based Arduino Nano, and a 3D-printed light-shielding enclosure. The signal pipeline uses lower-envelope detrending and zero-phase Butterworth filtering. HR is obtained by peak detection, SpO2 by the red/infrared ratio of ratios, and cuffless BP by first-derivative timing and amplitude features. For glucose, autoregressive, Teager–Kaiser energy, spectral-entropy, log-energy, and single-pulse morphological features were supplied to Linear Regression (LR), K-Nearest Neighbors (KNN), and Random Forest (RF) regressors. Data came from 30 fasting volunteers (32 recordings of 30 s each) with fingerstick glucometer, oscillometric cuff, and commercial pulse-oximeter references. On this dataset RF gave the lowest error (RMSE 11.11 mg/dL, MAE 7.85 mg/dL, R² = 0.8655) and 74.19% three-band classification accuracy, versus RMSE 24.31 and 26.45 mg/dL for LR and KNN. HR and SpO2 agreed closely with the references, and a single resting BP trial differed from the cuff by about 1.4 mmHg. Because the cohort is small, validation was limited, and glucose may be confounded by correlated physiological variables, these results demonstrate feasibility only. Subject-independent validation, Clarke/Parkes error-grid analysis, and a larger multi-day study are required before any clinical claim can be made.
No takes yet. Share an insight, caveat, or question.
Noman et al. (2026) conducted an observational in Healthy-to-mixed fasting cohort undergoing non-invasive multiparameter physiological monitoring (blood glucose, blood pressure, heart rate, SpO2) (n=30). Fingertip photoplethysmogram (PPG) prototype with Random Forest regressor vs. Linear Regression (LR), K-Nearest Neighbors (KNN), and commercial reference devices (Accu-Chek Active glucometer, oscillometric cuff, pulse oximeter) was evaluated on Blood glucose estimation accuracy (Root Mean Square Error [RMSE], Mean Absolute Error [MAE], R², and three-band classification accuracy) (RMSE 11.11 mg/dL, MAE 7.85 mg/dL, R² = 0.8655). A Random Forest regressor using fingertip PPG features estimated fasting blood glucose with an RMSE of /dL and 74.19% band accuracy, outperforming Linear Regression and KNN.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: