Name
Title | ||
|---|---|---|
Accuracy of a novel Raman-based technology for the early detection of multidrug-resistant bacteria |
Abstract | ||
|---|---|---|
Cultural methods, although time consuming, are still the gold standard for the microbial detection, combining high sensitivity and specificity. Vibrational spectroscopies, such as Raman spectroscopy, have been recently proposed as an alternative, being label-free, non-invasive, and highly specific. This study aimed to evaluate the accuracy of a new technology (AMR-S3DP, Sens Solutions, Barcelona, Spain) based on Raman spectroscopy to detect the presence of three clinically relevant multidrug resistant pathogens (Clostridium difficile, Klebsiella pneumoniae and methicillin resistant Staphylococcus aureus). Different machine learning strategies for analyzing the collected Raman spectra were compared to find a solution trading-off between accuracy and computational cost. Experimental datasets were collected in controlled conditions using pure cultures of the three microorganisms of interest. Then, nine state-of-the-art classifiers and several instances of a Multi-Layer Perceptron Neural Network were trained and tested using the dataset. Three experiments were ran: (i) classification of only the three bacteria strains, (ii) classification of the three bacteria strains and the absence of bacteria, (iii) the same settings with standardized and normalized data. All the experiments were performed following a 10-Fold stratified Cross-validation approach. Tested methods included: Logistic regression, Nearest Neighbor Classifier, Support vector machines, Gaussian process, Decision Trees, Random Forest, Boosting, and Quadratic Classifier Naïve Bayes. Data distributions were highly nonlinear, nevertheless, Gaussian Process and Logistic Regression clearly outperformed the other tested methods when training and testing data sets were normalized and standardized. Gaussian Processes failed in providing a competitive solution to be executed in low-cost devices, being several orders of magnitude slower than Logistic Regression. With the most performant analytical method, a precision >94% and a recall rate >95% was obtained for all the three microorganisms of interest, making the system suitable to detect MDR pathogens and competitive with current gold standard techniques in term of time to detection. |
Year | DOI | Venue |
|---|---|---|
2022 | 10.1109/MeMeA54994.2022.9856540 | 2022 IEEE International Symposium on Medical Measurements and Applications (MeMeA) |
Keywords | DocType | ISBN |
MDRO,Clostridium difficile,Klebsiella Pneumoniae,MRSA,organic volatile compounds,hospital acquired infection,spectrometry,neural networks | Conference | 978-1-6654-8300-1 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
7 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Jacopo Dolcini | 1 | 0 | 0.34 |
| Sílvia Gómez-Montes | 2 | 0 | 0.34 |
| Raquel Obregón | 3 | 0 | 0.34 |
| Marck Collado | 4 | 0 | 0.34 |
| Francesco Barbabella | 5 | 0 | 0.34 |
| Carlos Chiatti | 6 | 0 | 0.34 |
| Francesco Tessarolo | 7 | 0 | 1.01 |