Sepsis presents a significant global health concern, uniquely challenging as current diagnostic tools are limited in timeliness in both testing and detection ability for early stages of the condition. Surface-enhanced Raman spectroscopy (SERS) has demonstrated great potential in sensitive and selective identification of biomarkers within human samples. However, the information density within a SERS spectrum poses a challenge for interpretation and analysis of a patient sample. Machine learning (ML) can be leveraged to learn patterns within the data and enable identification and categorization of critical biomarkers. This paper presents two SERS based ML classification models to rapidly identify and quantify sepsis severity even in the early stages of a condition, using low sample volume. The first model detects Ang-1, Ang-2, and sTREM-1 sepsis biomarkers within human plasma samples at clinically significant concentrations outside of the normal human range, with up to 83.5% accuracy overall and 100% accuracy on normal samples. The second model categorizes sTREM-1 samples into clinically determined concentration ranges associated with 3 stages of sepsis severity, with up to 100% accuracy. Insight is also provided on key wavenumber regions for discrimination between biomarkers and normal plasma. Our findings illustrate the feasibility of SERS based ML approaches for rapid and early detection of life-threatening sepsis scenarios.
扫码关注我们
求助内容:
应助结果提醒方式:
