Hybrid calibration algorithm based on RKHS-PLSR for high-accuracy noninvasive optical monitoring of blood parameters

During the clinical extracorporeal life support process, noninvasive optical measurements of hematocrit (HCT) and blood oxygen saturation (SO₂) often suffer from limited accuracy and a need for frequent calibration, particularly for HCT, and cannot yet provide truly continuous, high-accuracy monitoring. To address these limitations, a hybrid calibration algorithm combining the reproducing kernel Hilbert space (RKHS) with partial least squares regression (PLSR) is proposed. The algorithm first constructs a physically interpretable baseline model via the modified Lambert-Beer law and then integrates an RKHS-based anomaly suppression mechanism to inversely correct outlier interference in the original optical signals. This framework ultimately leverages the PLSR algorithm to conduct multiparameter joint calibrations that couple multiwavelength light intensity features with physical prediction outcomes. This hybrid framework effectively reduces the recalibration frequency while increasing the detection accuracy. A custom-designed multiwavelength inline optical monitoring system was developed and validated using blood samples with varying HCT and SO₂ levels. The experimental results demonstrate that, compared with conventional PLSR, the RKHS-PLSR algorithm reduces the root mean square errors (RMSE) of the HCT and SO₂ measurements to 1.41% and 1.23%, respectively. These results demonstrate that this innovative fusion algorithm, which synergizes physical correction with data calibration, significantly enhances the accuracy and long-term stability of continuous optical monitoring within extracorporeal circuits, reduces the recalibration frequency, and improves clinical reliability.