Mid-infrared multispectral lensless imaging for wide-field and label-free microbial identification

Abstract

Microbial identification is a critical process aiming at identifying the species contained in a biological sample, with applications in healthcare, industry or even national security. Traditionally, this process relies either on MALDI-TOF mass spectroscopy, on biochemical tests and on the observation of the morphology of colonies after growth on a Petri dish. Here is presented an innovative method for label-free optical identification of pathogens, based on the multispectral infrared imaging of colonies. This lensless imaging technique enables a high-throughput analysis and wide-field analysis of agar plates. It could yield very high correct identification rates as it relies on an optical fingerprint gathering both spectroscopic and morphologic features.The setup consists of a Quantum Cascade Lasers light source and an imager, a square 2.72 by 2.72 mm uncooled bolometer array. Microorganisms to be analyzed are streaked on a porous growth support compatible with infrared imaging, laid on top of an agar plate for incubation. When imaging is performed, growth support is put in close contact with the imaging sensor and illuminated at different wavelengths. After acquisition, an image descriptor based on spectral and morphological features is extracted for each microbial colony. Supervised classification is finally performed with a Support Vector Machine algorithm and tested with tenfold cross-validation. A first database collecting 1012 multispectral images of colonies belonging to five different species has already been acquired with this system, resulting in a correct identification rate of 92%. For these experiments, multispectral images are acquired at nine different wavelengths, between 5.6 and 8 µm.Considering the optimization possibilities of the image descriptors currently used and the ongoing development of the uncooled bolometers technology, these very first results are promising and could be dramatically improved with further experiments. Thereby, mid-infrared multispectral lensless imaging has the potential to become a fast and precise Petri dish analysis technology.

Simple imaging system for label-free identification of bacterial pathogens in resource-limited settings

“…Classification in further studies could possibly be improved through the acquisition of samples at different wavelengths. While combining image features acquired at two wavelengths (one in the visible range and another in the infrared one) did not yield a statistically significant gain for this study (see hyperparameter study in section S3.1), we believe such data enrichment to show potential, as the aspect of scatterograms wildly varies across wavelengths [Dupoy et al, 2020] (Figure S3). Altogether, we are confident that the coupling of an imaging system capable of generating intricate patterns with an algorithm capable of automatically extracting highly complex features is a promising avenue, and that more substantial datasets and improvements in the network's architecture could be sufficient to lead to sizeable performance boosts.…”

Ultrafast and Multiplexed Bacteriophage Susceptibility Testing by Surface Plasmon Resonance and Phase Imaging of Immobilized Phage Microarrays

“…Optical label-free techniques have been demonstrated to be effective for non-destructive characterization of bacterial pathogens, phenotypic identification [22] and viability assays [23], not only at the level of colonies [24] and microcolonies [25], but also down to the scale of single-cells [26]. In particular, imaging of phage plaques with wide-field lens-less techniques was recently reported to enable rapid solid-phase phage susceptibility testing with time to results as low as 4h20 for anti-Staphylococcus aureus phage and 2h20 for anti-Klebsiella pneumoniae phage [14].…”

Microbial identification through multispectral infrared imaging of colonies: combining chemical and morphotype analysis

“…In opposition to FTIR, DFIR operates by successively lighting the sample with a set of discrete wavelengths. Recent examples of applications include chemical imaging of human tissues [22][23][24], real-time imaging of microorganisms [25], or classi cation of several bacterial species [26,27]. QCLs are well suited for infrared imaging, and already offer a signi cant gain of time compared to FTIR imaging: a few minutes against hours, for the same surface with the same resolution [28].…”