A carregar...
Publicação
Rapid FTIR spectral fingerprinting of kidney allograft perfusion fluids distinguishes DCD from DBD donors: a pilot machine learning study
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.49 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Background/Objectives: Rapid, objective phenotyping of donor kidneys is needed to support peri-implant decisions. Label-free Fourier-transform infrared (FTIR) spectroscopy of static cold-storage Celsior® perfusion fluid can discriminate kidneys recovered from donation after circulatory death (DCD) versus donation after brain death (DBD). Methods: Preservation solution from isolated kidney allografts (n = 10; 5 DCD/5 DBD) matched on demographics was analyzed in the Amide I and fingerprint regions. Several spectral preprocessing steps were applied, and feature extraction was based on the Fast Correlation-Based Filter. Support vector machines and Naïve Bayes were evaluated. Unsupervised structure was assessed based on cosine distance, multidimensional scaling, and hierarchical clustering. Two-dimensional correlation spectroscopy (2D-COS) was used to examine band co-variation. Results: Donor cohorts were well balanced, except for higher terminal serum creatinine in DCD. Quality metrics were comparable, indicating no systematic technical bias. In Amide I, derivatives improved classification, but performance remained modest (e.g., second derivative with feature selection yielded an area under the curve (AUC) of 0.88 and an accuracy of 0.90 for support vector machines; Naïve Bayes reached an AUC of 0.92 with an accuracy of 0.70). The fingerprint window was most informative. Naïve Bayes with second derivative plus feature selection identified bands at ~1202, ~1203, ~1342, and ~1413 cm−1 and achieved an AUC of 1.00 and an accuracy of 1.00. Unsupervised analyses showed coherent grouping in the fingerprint region, and 2D correlation maps indicated coordinated multi-band changes. Conclusions: Performance in this 10-sample pilot should be interpreted cautiously, as perfect leave-one-out cross-validation (LOOCV) estimates are vulnerable to overfitting. The findings are preliminary and hypothesis-generating, and they require confirmation in larger, multicenter cohorts with a pre-registered analysis pipeline and external validation.
Descrição
This research was funded by Centro Clínico Académico de Lisboa, grant number FFCCAL.
05.2025.
Palavras-chave
FTIR spectroscopy Kidney transplantation Perfusion fluid DCD vs. DBD Machine learning
Contexto Educativo
Citação
Ramalhete, L., Araújo, R., Vieira, M. B., Vigia, E., Pena, A., Carrelha, S., Ferreira, A., & Calado, C. R. C. (2025). Rapid FTIR spectral fingerprinting of kidney allograft perfusion fluids distinguishes DCD from DBD donors: A pilot machine learning study. Metabolites, 15(11), 702. https://doi.org/10.3390/metabo15110702
Editora
MDPI AG