HCAIs and Detection Challenges
Healthcare-associated infections (HCAIs), like respiratory, bloodstream, and urinary tract infections, are emerging as a major contributor to global mortality owing to rising antimicrobial resistance. The primary causative pathogens of HCAIs include Escherichia coli (E. coli) and Staphylococcus aureus (S. aureus). Thus, accurate and faster detection of causative pathogens is crucial for reducing HCAI mortality and incidence.
This has resulted in increased attention on high-throughput, rapid bacterial detection technologies to address antimicrobial resistance and promptly enable targeted antimicrobial therapy. Traditional bacterial detection approaches have several drawbacks, like costly equipment and a lack of high sensitivity. They are also unsuitable for point-of-care applications.
VOC Profiling and PID Advances
Bacteria emit volatile organic compounds (VOCs), like sulfur-containing compounds, terpenoids, and alcohols during their metabolism, with each species having a specific VOC fingerprint. Thus, VOC profiling could be a suitable technique for detecting and identifying bacteria.
However, the detection of bacterial VOCs using metal oxide sensors (MOS), gas chromatography–mass spectrometers (GC-MS), or electronic noses remains challenging. For instance, GC-MS is expensive, time-consuming, and lacks real-time monitoring ability.
PID relies on vacuum ultraviolet (VUV) photon emission by a light source having a specific energy. Every VOC possesses a distinct threshold molecular ionization energy, or ionization potential (IP). Hence, the PID ionizes the VOC when the VUV photon energy surpasses the VOC IP. Recent advances in PID technology have enabled portable, low-cost, and humidity- and temperature-resistant PIDs.
While VOC profiling studies have used dimensionality reduction and clustering techniques, only a few have explored supervised machine learning (ML) techniques. Lack of adequate data is a major limitation in effectively training such models.
The Proposed Solution
In this work, researchers introduced a real-time multiplex lamp PID assisted by AI image-based analysis for bacterial species identification. In the PID, four lamps with varying ionization energies were used to obtain four distinct current curves in real time for each target bacterial species, including S. aureus, E. coli, Klebsiella pneumoniae (K. pneumoniae), and Pseudomonas aeruginosa (P. aeruginosa).
The sensing approach focused on monitoring the signal’s temporal response pattern, which was later analyzed using AI techniques to obtain diagnostic information. In this study, the PID-based system directly measured the dynamic response of overall bacterial VOC emissions in real time in the headspace, without relying on chromatographic separation, analyte-specific extraction, or compound identification.
This allowed direct bacterial detection from a culture bottle with no VOC pre-concentration or sample preparation, which is required for GC-MS. Thus, pre-enrichment steps like time-consuming spectral acquisitions were eliminated.
Researchers transformed the current curves into image representations to capture their distinct patterns for bacterial differentiation. Specifically, a pre-trained ResNet-18 convolutional neural network (CNN) was used within a few-shot learning (FSL) framework.
They selected the FSL strategy to achieve accurate classification even with fewer labeled examples for each class. FSL acts as a practical framework to investigate the automated discrimination feasibility under constrained conditions in the initial AI implementation stage, where data acquisition is expensive, and availability is limited. FSL’s application also establishes a methodological base for later development stages.
Hence, the PID + AI approach proposed here could enable label-free VOC detection without sample preparation and serve as a simple sensing system following a more streamlined approach for rapid bacterial detection. Another crucial advantage was its versatility across diverse background matrices, enabling accurate bacterial identification provided that bacterial VOC patterns remained distinct from the surrounding medium, even in complex samples like blood or urine.
Importance of this Work
Researchers distinguished four bacterial species based on distinct current signals generated by four PID sensors, reflecting their different metabolic signatures. Compared with GC-MS, the proposed system operated faster and was smaller.
It was more compact and portable, with a total volume of about 0.028 m³, with further miniaturization allowed. The similar design of electronics, lamp chambers, and lamp drivers simplified construction, reduced manufacturing complexity, and lowered cost.
Additionally, the sensor demonstrated high sensitivity, enabling real-time headspace analysis of bacterial cultures without pre-concentration and detecting levels as low as 10² CFU/mL, which was clinically relevant. This detection level aligned with HCAI ranges, including bloodstream infections (1–200 CFU) and urinary tract infections ~105 CFU.
A pre-trained ResNet-18 CNN within an FSL framework achieved over 88% accuracy in bacterial species differentiation under limited labeled data. The integration of PID sensing with AI-based analysis provided a rapid diagnostic approach with strong clinical potential.
In conclusion, the findings of this study demonstrated the feasibility of the proof-of-concept sensor as a rapid, early-detection tool for suspected infections.
Saving this for later? Download a PDF here.
Journal Reference
Costa, S. P. et al. (2026). Bacterial species differentiation via real-time detection of microbial volatile organic compounds using a wavelength multiplexed photoionization detector and AI image-based analysis. Scientific Reports. DOI: 10.1038/s41598-026-46818-x, https://www.nature.com/articles/s41598-026-46818-x#citeas
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.