Extracorporeal membrane oxygenation (ECMO) is a medical device that provides temporary external circulation and respiratory support during heart-lung procedures, substituting for heart and lung function to alleviate their burden and allow more time for treatment. This study employs the sequential organ failure assessment (SOFA) to evaluate the severity of illness in ECMO patients and utilizes noninvasive near-infrared spectroscopy (NIRS) to monitor lower limb microcirculation. By extracting and selecting features, blood oxygen information is input into machine learning models for classification and regression analysis. The results indicated that the classification accuracy for disease severity reached 90% for veno-venous (VV-ECMO) and veno-arterial (VA-ECMO) patients, demonstrating the efficacy of combining NIRS with machine learning in clinically distinguishing disease severity. Additionally, the regression analysis yielded excellent performance. These findings underscore the effectiveness of NIRS in assessing disease severity among ECMO patients, offering valuable clinical guidance for optimizing ECMO settings, adjusting cardiovascular medication dosages, and predicting patient prognosis.