1372 Non-Invasive Diagnosis of Moyamoya Disease Using Serum Metabolic Fingerprints and Machine Learning

INTRODUCTION: Moyamoya disease (MMD) is a progressive cerebrovascular disorder that increases the risk of intracranial ischemia and hemorrhage. Timely diagnosis and intervention can significantly reduce the risk of new-onset stroke in patients with MMD. However, the current diagnostic methods are invasive and expensive, and non-invasive diagnosis using biomarkers of MMD is rarely reported. METHODS: In our study, we employed nanoparticle-assisted LDI MS for the extraction of serum metabolic fingerprints (SMFs) from MMD group, aiming for non-invasive diagnosis.Utilizing machine learning, we distinguished the MMD group from the HC group. A cluster analysis was performed on the group of MMD patients based on metabolites to examine the variations in cognitive impairment levels across groups. RESULTS: In our study, we analyzed 288 serum samples from 144 Moyamoya disease (MMD) patients and 144 healthy controls (HC), demonstrating our deep learning-based diagnostic model's high efficacy in distinguishing MMD from HC, with an AUC of 0.977. We identified six metabolites significantly altered between the groups, indicating a potential MMD-specific metabolite panel. Additionally, hierarchical clustering of these metabolites revealed three distinct cognitive patterns correlating with MMD severity, suggesting their clinical relevance in understanding and diagnosing MMD. CONCLUSIONS: we achieved rapid sample detection using the nanoparticle-assisted LDI MS platform. Coupled with machine learning, we successfully diagnosed MMD patients with an AUC of up to 0.977. We discovered a strong correlation between specific MMD metabolite patterns, neuropsychological performance, and the incidence of VCI.