931. AI-Enhanced Image-Based Bacterial Differentiation via Volatile Compound Detection
Invited abstract in session TC-13: AI in healthcare, stream OR in Healthcare (ORAHS).
Tuesday, 12:30-14:00Room: Clarendon SR 1.01
Authors (first author is the speaker)
| 1. | António Cardoso
|
| CEGI, INESC TEC | |
| 2. | Felipe Tetsuo Yamada
|
| INESC TEC, Faculdade de Engenharia, Universidade do Porto | |
| 3. | Flávia Barbosa
|
| Porto, INESC TEC | |
| 4. | Luis Guimarães
|
| INESC TEC, Faculadade de Engenharia, Universidade do Porto |
Abstract
Healthcare-associated infections (HCAIs) represent a serious global health threat exacerbated by antimicrobial resistance. Traditional detection methods, such as culturing and molecular techniques, have limitations, including low sensitivity and high costs. An alternative strategy is the detection of volatile organic compounds (VOCs) emitted by bacteria, which serve as a unique fingerprint for bacterial identification.
This study presents a novel bacterial detection method using a wavelength-multiplexed photoionization detector (PID) to generate distinct ionization current patterns specific to different bacteria based on the VOCs they emit. These patterns are transformed into image representations, which are then analysed using Few-Shot Learning (FSL), and a machine learning approach designed to work effectively with limited labelled data.
To extract robust features from the image-based data, a pre-trained Convolutional Neural Network (CNN), specifically ResNet-18, is employed. Prototypical Networks (PN) are then applied to classify bacterial species by comparing query samples to prototypes representing each class. This AI-driven methodology offers a promising solution for real-time bacterial detection, particularly in clinical diagnostics and infection control, where limited data poses a significant challenge.
Keywords
- Artificial Intelligence
- Machine Learning
- Algorithms
Status: accepted
Back to the list of papers