Soft, Multi-Wavelength Photoplethysmography Enables Reliable Neonatal Blood Pressure Monitoring Via Error Stratification.
Continuous noninvasive blood pressure monitoring is important for neonatal hemodynamic management. However, cuffless photoplethysmography (PPG) is still difficult to use clinically because acquisition-related errors and their effects on blood pressure estimation are not well understood. Most studies report only overall error, which can hide error patterns under different signal-quality and motion conditions. Here, we developed a soft multi-wavelength wearable that combines reflective and transmissive optical paths to record complementary PPG signals. We also built NEO-BP, a prospective clinical dataset containing synchronized multi-wavelength PPG and invasive arterial blood pressure waveforms from 42 neonates. In a held-out segment-level test set without excluding samples based on signal quality (n = 9378), the green-red-infrared (G+R+IR) model achieved mean absolute errors of 9.67 mmHg for systolic blood pressure and 6.17 mmHg for diastolic blood pressure. Using this dataset, we established a retrospective descriptive subgroup error profile by stratifying estimation deviations across measurable acquisition and physiological conditions. Estimation errors increased under lower signal quality and stronger motion. These results identify error-prone acquisition states, support quality-aware interpretation of segment-level cuffless blood pressure estimates, and provide a basis for future models that estimate the error risk of individual readings.