講演情報
[8a-A22-7]SERS based Explainable Machine Learning Framework for Patient-Independent Detection of Respiratory Viral Co-Infections
〇(DC)Rakesh Naik1, Arti Yadav1, Ekta Gupta2, Sachin Kumar Srivastava1 (1.IIT Roorkee, 2.ILBS Delhi)
キーワード：
Surface enhanced Raman Spectroscopy、Machine learning、Respiratory viral co-infection
Respiratory viral co-infections are associated with increased disease severity and overlapping clinical symptoms. This makes rapid diagnosis challenging. Surface-enhanced Raman spectroscopy (SERS) offers a label-free approach for probing infection-associated biochemical signatures with high sensitivity. However, simultaneously identifying multiple pathogens while preserving interpretability and patient independence remains difficult. Here we propose a SERS based explainable machine learning framework for patient-independent screening and identification of respiratory viral co-infections in two stages. Stage 1 employs a binary XGBoost classifier, trained on patient-independently partitioned spectra, to discriminate negative from infection-associated SERS signatures, achieving 98% sensitivity, 89% specificity, and an ROC–AUC of 0.97. Stage 2 performs multi-label virus identification using a MultiOutput XGBoost model trained on synthetic one to five-virus different virus co-infection spectra generated from patient-independent libraries of Delta, Omicron, Influenza, Rhinovirus, and Respiratory Syncital Virus (RSV), achieving micro and macro-averaged F1-scores of 0.81. SHAP analysis reveals overlapping infection-associated bands between the two stages, suggesting a shared biochemical basis for screening and virus-specific refinement. This work demonstrates strong potential for explainable point-of-care surveillance.