SERS on analyte-enriched blood for rapid, culture-free sepsis recognition and causative pathogen identification with super operational neural networks
<p dir="ltr">Sepsis remains a leading cause of morbidity and mortality, yet routine diagnostics are slow, culture-dependent, and often lack the sensitivity or specificity required for early intervention. Prior studies rarely demonstrate clinical-grade performance on blood culture sam...
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| مؤلفون آخرون: | , , , , , , , , , , , , |
| منشور في: |
2025
|
| الموضوعات: | |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | <p dir="ltr">Sepsis remains a leading cause of morbidity and mortality, yet routine diagnostics are slow, culture-dependent, and often lack the sensitivity or specificity required for early intervention. Prior studies rarely demonstrate clinical-grade performance on blood culture samples or in independent external cohorts. We address these gaps with a surface-enhanced Raman spectroscopy and deep learning workflow (SERS-DL) that performs sepsis instance recognition and causative pathogen identification directly from target-analyte enriched blood. We assembled a primary dataset of SERS spectra acquired from 653 analyte-enriched blood samples collected at a tertiary hospital in Qatar and an external blind cohort of 70 independent samples. After rigorous preprocessing and class-weighted augmentation of SERS spectra, we trained SuperRamanNet, a lightweight one-dimensional classifier based on super operational neural networks. In five-fold, sample-contained cross-validation, the system achieved 99.67 % accuracy for binary sepsis recognition and 98.84 % accuracy for six-class pathogen identification. On the external cohort, performance remained high at 98.28 % for pathogen typing, indicating robust generalizability. Comparative benchmarks and ablation studies confirmed consistent gains over convolutional and operational baselines and quantified the impact of augmentation and architectural choices. Residual confusions were concentrated between control and <i>Escherichia coli</i> and among certain Gram-negative classes, underscoring the need for improved raw class balance during blood sample collection. Overall, this rapid, culture-free, and portable SERS-DL pipeline delivers near clinical-grade accuracy for sepsis detection and pathogen identification directly from blood. The compact model and streamlined workflow support point-of-care translation, with potential to accelerate triage, guide early therapy, and reduce the global sepsis burden.</p><h2 dir="ltr">Other Information</h2><p dir="ltr">Published in: Talanta<br>License: <a href="http://creativecommons.org/licenses/by/4.0/" target="_blank">http://creativecommons.org/licenses/by/4.0/</a><br>See article on publisher's website: <a href="https://dx.doi.org/10.1016/j.talanta.2025.129332" target="_blank">https://dx.doi.org/10.1016/j.talanta.2025.129332</a></p> |
|---|