Early detection and classification of live bacteria using time-lapse coherent imaging and deep learning - PubMed (original) (raw)
Early detection and classification of live bacteria using time-lapse coherent imaging and deep learning
Hongda Wang et al. Light Sci Appl. 2020.
Abstract
Early identification of pathogenic bacteria in food, water, and bodily fluids is very important and yet challenging, owing to sample complexities and large sample volumes that need to be rapidly screened. Existing screening methods based on plate counting or molecular analysis present various tradeoffs with regard to the detection time, accuracy/sensitivity, cost, and sample preparation complexity. Here, we present a computational live bacteria detection system that periodically captures coherent microscopy images of bacterial growth inside a 60-mm-diameter agar plate and analyses these time-lapsed holograms using deep neural networks for the rapid detection of bacterial growth and the classification of the corresponding species. The performance of our system was demonstrated by the rapid detection of Escherichia coli and total coliform bacteria (i.e., Klebsiella aerogenes and Klebsiella pneumoniae subsp_. pneumoniae_) in water samples, shortening the detection time by >12 h compared to the Environmental Protection Agency (EPA)-approved methods. Using the preincubation of samples in growth media, our system achieved a limit of detection (LOD) of ~1 colony forming unit (CFU)/L in ≤9 h of total test time. This platform is highly cost-effective (~$0.6/test) and has high-throughput with a scanning speed of 24 cm2/min over the entire plate surface, making it highly suitable for integration with the existing methods currently used for bacteria detection on agar plates. Powered by deep learning, this automated and cost-effective live bacteria detection platform can be transformative for a wide range of applications in microbiology by significantly reducing the detection time and automating the identification of colonies without labelling or the need for an expert.
Keywords: Biophotonics; Imaging and sensing; Microscopy.
© The Author(s) 2020.
Conflict of interest statement
Conflict of interestH.C.K, H.W., Y.R., Y.Q., and A.O. have a patent application on the invention reported in this paper.
Figures
Fig. 1. High-throughput bacterial colony growth detection and classification system.
a Schematic of the device. b Photograph of the lens-free imaging system. c Detailed illustration of various components of the system
Fig. 2. Schematics demonstrating the workflow of the microorganism monitoring system.
a Bacterial sample preparation workflow. b Steps of the image and data processing algorithms for the automated detection of the growing colonies and classification of their species. The scale bars for the holographic images of the growing colonies (E. coli and K. aerogenes) and a static particle (dust) are 100 µm
Fig. 3. Images captured using the microorganism monitoring system.
a Whole agar plate image of mixed E. coli and K. aerogenes colonies after 23.5 h of incubation. b Example images (i.e., amplitude and phase) of the individual growing colonies detected by a trained deep neural network. The time points of detection and classification of growing colonies are annotated with blue arrows. The scale bar is 100 µm
Fig. 4. Sensitivity and precision analysis.
Sensitivity of growing colony detection using our trained neural network for aK. pneumoniae, bE. coli, and cK. aerogenes. d Precision of growing colony detection using our trained neural network for all three species. The pink arrow indicates the time for late “wake-up” behaviour for some of the E. coli colonies. e Characterizing the growth speed of chlorine-stressed E. coli using our system. There was an ~2 h delay in colony formation for chlorine-stressed E. coli (orange curve) compared to the unstressed E. coli strain (blue curve). The error bars show the standard deviation values across multiple plates
Fig. 5. Classification analysis.
Classification performance of our trained neural network for aK. pneumoniae, bE. coli, and cK. aerogenes colonies. The green shaded area in each curve represents the highest and lowest recovery rates found in all the corresponding experiments at each time point. d The blind testing confusion matrix of classifying all the colonies that were sent to our trained neural network after 12h of incubation. A diagonal entry of 1.0 means a 100% classification accuracy for that species. The numbers of colonies that were tested by the classification network in d are 325 (E. coli), 334 (K. pneumoniae), and 256 (K. aerogenes)
Fig. 6. Quantification of the LOD of our system.
a The CFU count from our system is plotted against the CFU/L counts of the spiked samples, calculated independently using the Colilert®-18 method after 18 h of incubation. CFU counts acquired with our platform at different time points are coloured from blue to yellow, which corresponds to 5–14.5 h of total test time, including the signal amplification step that involves liquid culture media (5 h). b Without signal amplification, the LOD is decreased due to the low transfer rate from the filter membrane to the agar surface (see Supplementary Figs. S3 and S4). c As a control experiment, we prepared and imaged 3 agar plates that showed <1 CFU count from our setup throughout the test period from 5 to 14.5 h. d The LOD of our system is ~11 CFU/L at 8.5 h and ~1 CFU/L at ≤9 h
References
LinkOut - more resources
Full Text Sources
Other Literature Sources
Research Materials