Introduction of research team and methods of Multi-modal integrated diagnosis of central nervous system infections.
Team
Prof. Longjiang Fan
fanlj@zju.edu.cn- Executive Director, Institute of Bioinformatics, Zhejiang University
- PI, Air Pollution and Health Research Center, Zhejiang University
- Professor, Department of Medical Oncology, the First Affiliated Hospital, Zhejiang University
Prof. Yu Chen
chenyuzy@zju.edu.cn- Professor, Centre of Clinical Laboratory, First Affiliated Hospital, Zhejiang University
- Director, Key Laboratory of Clinical In Vitro Diagnostic Techniques of Zhejiang Province, Zhejiang University
- Director, Institute of Laboratory Medicine, Zhejiang University
- PI, State Key Laboratory for Diagnosis and Treatment of Infectious Diseases, National Clinical Research Center for Infectious Diseases
Associate Prof. Yifei Shen
yifeishen@zju.edu.cn- Associate Professor, Centre of Clinical Laboratory, First Affiliated Hospital, Zhejiang University
- PI, Key Laboratory of Clinical In Vitro Diagnostic Techniques of Zhejiang Province, Zhejiang University
- PI, Institute of Laboratory Medicine, Zhejiang University
- Department of Bioinformatics and Computational Biology, The University of Texas MD Anderson Cancer Center
Methods
Current diagnostics for central nervous system infections (CNSIs)—including culture, serology, and targeted PCR—lack sensitivity and often fail to identify pathogens. Metagenomic next-generation sequencing (mNGS) improves detection breadth but remains limited by pathogen nucleic acid load and turnaround time. Integrating host response markers and routine lab data with mNGS may enhance diagnostic accuracy and clinical utility. We conducted a multicenter retrospective study involving 1,706 patients from five Chinese medical centers. We developed a multimodal machine learning model, InfectDiagno-CNS-fusion, integrating CSF host gene expression profiles, routine blood and CSF indices, and mNGS pathogen data. The model was trained and validated using ensemble machine learning algorithms. Diagnostic performance was evaluated through receiver operating characteristic (ROC) curves and cross-validation.