Research on Real-Time Monitoring of Parameters in Ex Vivo Perfused Red Blood Cells Based on Spectral Analysis Method

Ex vivo perfusion plays a critical role in organ preservation, transplantation, and clinical research. Real-time periodic monitoring of blood oxygen levels and hematocrit (HCT) is essential during ex vivo perfusion. Unlike in vivo monitoring, ex vivo perfusion necessitates long-term surveillance of red blood cell parameters under nonpulsatile conditions, which current in vivo devices are unable to achieve effectively. To address this gap, we have developed a portable, real-time, noninvasive blood gas monitoring system based on near-infrared spectroscopy (NIRS). The system comprises a portable near-infrared (NIR) spectrometer module and a PC-based predictive model. This study integrates field-programmable gate array (FPGA) technology with multiframe fusion and temporal filtering techniques to efficiently process the single-pixel signals of the spectra, thereby mitigating the jelly effect encountered during the acquisition of dynamic substances. This approach effectively reduces interference caused by pumps, blood flow, and dark current. In addition, Savitzky–Golay (S–G) smoothing was applied to the entire spectrum, enhancing the accuracy and stability of the data. Principal component analysis (PCA) was employed to identify the primary components in the dataset, and particle swarm optimization (PSO) was utilized to optimize the extreme learning machine (ELM) model. Compared with other algorithms, the proposed model demonstrated superior prediction accuracy and algorithmic simplicity, particularly for spectral data with a limited number of features. Experimental results demonstrated that the correlation coefficients for blood oxygen, HCT, and pH in the test set were 0.95, 0.94, and 0.85, respectively, with corresponding standard deviations of 4.05, 2.98, and 0.13. These results are comparable to those of leading international commercial devices. The system has been successfully implemented and validated in engineering applications. In conclusion, this study presents a novel approach for real-time, noninvasive monitoring of blood gas-related parameters during ex vivo perfusion, achieved by integrating a spectroscopic sensor with machine learning and chemometric methods. Furthermore, the sensor is adaptable for use in other dynamic environments for quantitative analysis and measurement.