Surface-enhanced Raman spectroscopy for rapid sepsis recognition and pathogen identification from blood cultures using super operational neural networks

Sepsis, a critical medical emergency driven by a dysregulated host response to infection, remains a leading cause of global morbidity and mortality. Current diagnostic methods are slow, blood culture-dependent, and often lack sensitivity or specificity, delaying timely intervention and contributing to poor outcomes. Recent advances in surface-enhanced Raman spectroscopy (SERS) and artificial intelligence (AI) offer promising solutions. Yet, existing machine learning studies have either failed to achieve clinical-grade performance or have not directly targeted rapid sepsis detection from blood cultures. In this study, we collected an extensive set of blood culture samples from a diverse patient cohort attended tertiary level hospital in Qatar, including both clinically confirmed sepsis-positive and control cases, then constructed a large SERS spectral dataset with additional external validation from an independent cohort. We propose SuperRamanNet, a novel deep learning framework based on lightweight, one-dimensional super generative neuron operational neural networks (Super-ONNs), for rapid sepsis recognition and multiclass pathogen identification directly from SERS spectra. The system demonstrates robust performance, achieving 99.67% accuracy for sepsis recognition and 98.84% accuracy for pathogen identification on the primary dataset, with similarly high results on external validation. Comparative analysis confirms that SuperRamanNet consistently outperforms benchmark models and previous literature, supported by ablation studies highlighting the impact of data augmentation and architectural innovations. In conclusion, this work establishes SuperRamanNet as a clinically viable, high-throughput, and portable diagnostic tool, capable of transforming sepsis detection and pathogen identification at the point of care and potentially reducing the global burden of sepsis.