What is it about?
Dangerous bacteria and fungi often hide together in sticky layers called biofilms on surfaces like hospital equipment, making infections hard to treat. Current detection methods are slow and require taking physical samples. We have developed a new technique using a special camera that captures light beyond what the human eye can see, combined with artificial intelligence. This allows us to accurately identify specific dangerous pathogens within mixed communities instantly, without touching or damaging the surface. This technology could enable rapid scanning of large areas to prevent outbreaks in healthcare and food safety settings.
Featured Image
Photo by Steve Johnson on Unsplash
Why is it important?
This work demonstrates, for the first time, that hyperspectral imaging combined with deep learning can accurately identify opportunistic pathogens—including drug-resistant ESKAPEE bacteria and Candida albicans—within mixed biofilm communities without physical contact or sample preparation. Two significant advances are that: a) structural and textural features captured in full 3D hyperspectral cubes contribute more to identification accuracy than extended spectral ranges alone, enabling cost-effective visible-band implementations; and b) a multi-label neural network architecture successfully detects arbitrary species combinations in polymicrobial biofilms, moving beyond the simplified monoculture scenarios that dominate prior literature. These findings are timely given the global antimicrobial resistance crisis and the urgent need for rapid, scalable environmental surveillance tools. By enabling non-contact, macroscale scanning of surfaces in healthcare, food safety, and water infrastructure settings, this approach could transform routine contamination monitoring from a slow, lab-dependent process into an immediate, field-deployable capability. The publicly available dataset and reproducible code further lower the barrier for adoption and extension by the broader research community, accelerating translation toward real-world infection prevention strategies.
Perspectives
Writing this paper was personally meaningful because it represents a convergence of three fields I am deeply passionate about: optical sensing, microbiology, and machine learning. When we first started exploring whether hyperspectral imaging could distinguish pathogens in mixed biofilms, many colleagues were skeptical—biofilms are notoriously complex, polymicrobial communities are harder to model than monocultures, and combining VNIR and SWIR data with deformable registration added layers of technical difficulty. Yet, seeing the CNN learn to extract discriminative features directly from 3D hypercubes, without manual feature engineering, was a genuinely exciting moment that reinforced my belief in letting data-driven methods reveal patterns we might not anticipate. What motivates me most is the potential for this work to move beyond the lab. The idea that a camera-based system could one day scan hospital surfaces, food processing equipment, or water infrastructure in real time—without swabs, without culturing, without delaying action—is what makes the long hours of calibration, validation, and debugging worthwhile. I hope this article inspires other researchers at the intersection of engineering and life sciences to tackle "messy" real-world problems, even when simplified models are easier to publish. Finally, I am grateful for the collaborative spirit of this project. Working with microbiologists who provided clinical isolates and biological insight, with optical engineers who optimized acquisition protocols, and with data scientists who refined the ML pipelines made this work possible. If this paper encourages more cross-disciplinary conversations—or if it helps someone working on infection prevention feel that rapid, non-contact detection is within reach—then it has achieved what I hoped for.
Dr. Aleksandr Sinitca
Sankt-Peterburgskij gosudarstvennyj elektrotehniceskij universitet LETI
Read the Original
This page is a summary of: Non-contact identification of opportunistic pathogens in mixed biofilm contaminations by hyperspectral imaging, Analytica Chimica Acta, February 2026, Elsevier,
DOI: 10.1016/j.aca.2026.345098.
You can read the full text:
Resources
Contributors
The following have contributed to this page
