Abstract
Because Bacillus anthracis is one of the most lethal bioweapons, it is critical to create rapid, label-free screening and early warning systems to detect and classify anomalies in bacillus form vegetative cell and spore concentrations in the air. Even though significant effort has been invested in the development of various sensor solutions to detect, monitor, and identify airborne biological agents, no standard, interoperable, real-time or near-real-time optical sensor-based biothreat monitoring solution exists. Aside from the numerous advantages of genomic methods in microbe identification, optical sensors and microscopy-based technologies provide advantages in terms of rapid detection and classification capabilities. The AI-powered biothreat detection software platform from DataSenseLabs can perform intermethod comparison to cross-validate the results acquired by various quantitative phase imaging (QPI) measurement methodologies. This platform feature—support for multisensory data input—is not merely the foundation of the R&D level cross-validation approach, but also the key component of interoperable verification of air sample content in the case of airborne biothreat. Depending on the study design, sample type, and light microscopic or QPI measurement method, the platform’s algorithm system can detect and monitor abnormalities in the concentration of bacillus form objects taken from the air with greater than 80–95% accuracy. Another goal of the platform is to serve as a standardized tool for biomedical, environmental, and CBRN scientists to train and validate their concepts in pathogen detection and classification use cases, allowing them to better understand the gaps and challenges associated with artificial intelligence-powered optical sensor systems.
You have full access to this open access chapter, Download chapter PDF
Keywords
- Artificial intelligence
- Deep learning
- Bacteria
- Bacillus anthracis
- Biothreat
- Bioweapon
- Object detection
- Computer vision
- Microscopy
- Quantitative phase imaging
- Cross-validation platform
- Interoperable verification
- Air sampling
- Airborne
- Detection
- Classification
Introduction
An extensive amount of effort has been put into the development of different sensor solutions to detect, monitor, and identify airborne biological agents. The variety of methods behind the several sensor solutions cannot go unnoticed, but no standard and interoperable EU-wide approach is available to set the threshold for monitoring biothreat either outdoors or within critical infrastructure. Several research and development studies suggest bacteria are capable of surviving the aerosol transport and they can travel much longer distances between the hosts than we previously thought [2, 5, 8,9,10]. The sporulated form is not the only option for bacteria for airborne dissemination [6], which highlights the importance of using classical optical methods powered by AI-supported solutions for rapid monitoring and detection.
A recent study carried out by the Pentagon revealed several economic barriers of genomic analysis methods to be adopted in daily routine in terms of bioaerosol monitoring and biothreat detection [7] which shows that we are unprepared for real time, or at least frequent monitoring of airborne biocontent.
Regarding the currently available solutions for pathogen detection, there is a trade-off between time and accuracy. While the gold standards for genus and strain level identification of bacteria are still the different genomic methods, the classical optical methods like different forms of quantitative phase imaging microscopy powered by AI-supported solutions offer the possibility of rapid and automated detection of suspicious pathogens either in water or air-based samples.
One of the reasons why there is no existing standard, interoperable, and real-time or quasi-real-time, optical sensor-based biothreat monitoring solution is the lack of platforms capable of comparative verification, monitoring, and data archiving for traceable intermethod comparison and cross-validation.
There are several other reasons behind the difficulties of standardization and interoperability of optical sensor-based rapid biothreat monitoring and detection. We do not intend to cover the below-mentioned list in this short publication, but it is important to mention some of these factors to highlight the complexity of the problem:
-
The variety of air sampling methods and devices and the lack of knowledge regarding the limit of detection between the several air sampling solutions make it even more difficult to establish protocols for bioaerosol monitoring [1, 3]. The current solution is to focus on use case-specific applications and to comply with the generally well-established, referring NATO (STANAG) and ISO standards regarding the overall sampled volume of the air to avoid statistical down-sampling.
-
The lack of widely accepted air quality standards [3] makes it very difficult to determine the exact baseline of the “background noise” within the air [4, 8, 11]—referring to the nonbiological particle components—in order to optimize and finetune the AI-supported solutions for pathogen detection.
-
The currently accepted standard pathogen detection and identification protocols are based on culturing following the sample collection which significantly elongates the lengths of the process [3] and also increases the methodological diversity between the existing protocols.
Even if some AI-supported optical sensor-based platform would exist, there are no standard testing and validation protocols to evaluate the application-specific performance metrics. This is the topic we are focusing on in the framework of this publication. We introduce the recent results regarding the performance of the “DataSenseLabs AI-supported biothreat detection platform” using two different approaches: field data (air sample) collection and computer simulation-based testing related to our cross-validation-based development strategy.
Since the disease control authorities are highly vigilant regarding the environmental presence of the most dangerous and unfortunately well-known member of the Bacillus cereus group [4, 12, 13], and it is very easy to access all the necessary components to create a virulent Bacillus anthracis strain, our AI-supported biothreat detection platform is currently being finetuned for the detection of bacillus form objects sampled from the air.
Methods and Results
Regarding the process flow of biothreat detection, “DataSenseLabs AI-supported biothreat detection platform” is currently proceeded by three separate components of processes sequentially following each other: (1) air sample collection; (2) sample preparation; (3) light microscopic measurement; and the final step is the AI-supported pathogen detection. In the current phase of the R&D, the optical microscopic measurements were carried out by a digital holographic microscope (DHM-HoloZcan-EPro2) provided by the EU Horizon-supported HoloZcan project, and by a reference differential interference contrast (DIC) light microscope (NIKON Eclipse Ti2).
The performance of the DataSenseLabs AI-platform solution was finetuned and tested on three different sample types: laboratory-made mixture of different bacterial strains including Bacillus subtilis (ATCC 6533); environmental field samples collected by the Coriolis Compact air sampling device manufactured by Bertin Technologies; and computer-simulated holographic particle populations in the diameter (0.5 < d < 2 μm) and refractive index (R = 1.4) range compatible with the optical properties of bacteria. In the case of field samples, the air samples were dissolved in 1 ml physiological saline on site and were placed on standard microscopic slides for the DHM and DIC measurements.
Data analysis pipeline developed in MATLAB 2022b and Python consists of the following steps:
-
1.
Use case-specific database building from the captured microscopic image datasets. Currently, the DHM and DIC image data input is supported by the “DataSenseLabs AI-supported biothreat detection platform.”
-
2.
Extension of the databases with the metadata of annotated region of interests (ROIs). For supervised deep learning algorithm development, the ROIs of bacterial cells—as the reference or “ground Truth”—were labeled by rectangular bounding boxes in the MATLAB Image Labeler app.
-
3.
Asking the right questions: dividing the database into context-specific and use case-specific training and test sets.
-
4.
Optimization of the convolutional neural network (CNN) model parameters by training and data augmentation.
-
(a)
Optimizing the training process by finding the optimal number of training iterations based on the mini-batch accuracy and loss function values.
-
(b)
Image minibatches were created by dividing the original raw images into several smaller images with the pixel dimensions of 300 × 340. The training process was conducted by applying overlapping and nonoverlapping minibatches as well. During the training process, the saturation and exposure of the input images were modified within a 10% range, to diversify the training set.
-
(a)
-
5.
Testing the trained model, comparing the reference (ground Truth) and predicted values and metrics.
-
6.
Concluding the results and restart the process either from point 3 or 2 or even from point 1.
To support the original interoperability and cross-validation concept of the “DataSenseLabs AI-supported biothreat detection platform,” the possibility of multidata input has been developed to compare the results of the AI-supported detection in the case of several light microscopic image databases. In this publication, we show the summary statistics only of the DHM and DIC image datasets.
The two figures below (Fig. 10.1a and b) explain the concept of the cross-reference database-building strategy for the AI-supported platform development:
As referred to in Fig. 10.1a, we followed a custom-designed database-building strategy during the research and development process of the “DataSenseLabs AI-supported biothreat detection platform.” We have built seasonal, and geolocation-specific multiimage databases captured from the air samples to be able to estimate the overall diameter distribution of the naturally occurring biological and nonbiological particle components (“background noise”), since these can be crucial carriers of long-distance bacterial dissemination [4, 6, 8, 10, 11]. Three different approaches have been worked out to develop the databases for AI algorithm development and testing:
-
Calibration-oriented databases including the diameter values (in μm) of the seasonally occurring particles in the air are sampled by a reference air sampling device, the Coriolis Compact. The samples were taken at the same geolocations (forest region, busy road, indoors) in three different seasons (Summer, Autumn, and Spring).
-
Digitally mixed databases containing the microscopic images of the environmental (background noise) samples and the samples created under laboratory conditions including specific bacterial strains.
-
Directly mixed databases where the environmental fluid samples were directly mixed with the suspensions of specific, known bacterial strains.
The air samples were analyzed by DIC light microscopy to create the reference diameter value databases to verify the presence of the Bacillus simulant applied during the field data collection (air sample collection) and testing events as referenced by the figure below (Fig. 10.2).
Following the calibration-level verification of the presence of the artificially administered Bacillus simulants, the “DataSenseLabs AI-supported biothreat detection platform” was used to detect the presence of suspicious bacillus form objects within the air sample. The performance metrics are summarized below in Table 10.1a and Table 10.1b:
Besides the field data collection and testing-based AI algorithm development approach, it is extremely important to test and validate the theoretical limits of performance of an AI-supported algorithm. Simulated holographic particle databases have been created by the Mie method applying the Python-based HoloPy tool [14] to evaluate the performance of the “DataSenseLabs AI-supported biodetection platform” under use case-specific conditions as summarized in Fig. 10.3 and Table 10.2 below.
Conclusion and Further Steps
The “DataSenseLabs AI-supported biodetection platform” supports the quantitative phase imaging light microscopic sensor-based data input (DHM, DIC) and also two different computer-simulated image data inputs for algorithm development, training, testing, and evaluation. The platform’s algorithm system can detect and monitor the anomalies in the concentration of bacillus form objects sampled from the air with higher than 80–95% accuracy depending on the study design, sample type, and the light microscopic measurement method.
Evaluation of an AI-supported computer vision platform needs standardized, classic image databases as it is also suggested in the ITU-T F.748.12 (Table 7–3, page 4–5) standard [15], but we also suggest the implementation of a custom-designed, cross-validation-based three step (laboratory-made, field-collected, and simulated data-based) evaluation strategy as we introduced in the case of our AI platform development and evaluation. The simulated databases will be shared in a separate publication according to the “FAIR” data policy directives [16].
Based on the current results, the platform is capable of supporting CBRN and biothreat surveillance-related decision-making and prestandardization processes to establish solid foundations for interoperability in the field of optical sensor-based and light microscopic measurement-based instant biothreat detection. Following the analysis of the final results in the first part of 2024, we will be able to choose to most appropriate and use case-specific sensor type for miniaturization and industrial-level production.
References
Zhao, Y., Aarnink, A. J. A., Groot Koerkamp, P. W. G., et al. (2011). Detection of airborne Campylobacter with three bioaerosol samplers for alarming bacteria transmission in broilers. Biological Engineering, 3(4), 177–186.
Fernandez, M. O., Thomas, R. J., Garton, N. J., et al. (2019). Assessing the airborne survival of bacteria in populations of aerosol droplets with a novel technology. Journal of the Royal Society Interface, 16(150), 20180779.
Centers for Disease Control and Prevention. (2015). Background F. Environmental sampling. https://www.cdc.gov/infectioncontrol/guidelines/environmental/background/sampling.html. Accessed 30 July 2023.
Hagiwara, K., Matsumoto, T., Tsedendamba, P., et al. (2020). Distribution of viable bacteria in the dust-generating natural source area of the Gobi region, Mongolia. Atmosphere, 11(9), 893.
Clark, R. D., Underwood, G., McGenity, T. J., & Dumbrell, A. J. (2021). What drives study-dependent differences in distance–decay relationships of microbial communities? Global Ecology and Biogeography, 30(4), 811–825.
Hara, K., Zhang, D., Yamada, M., Matsusaki, H., & Arizono, K. (2011). A detection of airborne particles carrying viable bacteria in an urban atmosphere of Japan. Asian Journal of Atmospheric Environment, 5(3), 152–156.
Srikrishna, D. (2023). The world is completely unprepared for detecting emerging viruses/bacteria in the air. medRxiv. https://www.medrxiv.org/content/medrxiv/early/2023/03/29/2022.08.09.22278555.full.pdf. Accessed 30 July 2023.
Sanz, S., Olarte, C., Hidalgo-Sanz, R., Ruiz-Ripa, L., Fernández-Fernández, R., García-Vela, S., Martínez-Álvarez, S., & Torres, C. (2021). Airborne dissemination of bacteria (enterococci, staphylococci and Enterobacteriaceae) in a modern broiler farm and its environment. Animals, 11(6), 1783.
Zhao, J., Jin, L., Wu, D., et al. (2022). Global airborne bacterial community-interactions with Earth’s microbiomes and anthropogenic activities. Proceedings of the National Academy of Sciences of the United States of America, 119(42), e2204465119.
Woo, C., & Yamamoto, N. (2020). Falling bacterial communities from the atmosphere. Environmental Microbiomes, 15(1), 22.
Hu, W., Murata, K., Fan, C., Huang, S., Matsusaki, H., Fu, P., & Zhang, D. (2020). Abundance and viability of particle-attached and free-floating bacteria in dusty and nondusty air. Biogeosciences, 17, 4477–4487.
Centers for Disease Control and Prevention. (2020). Anthrax. https://www.cdc.gov/anthrax/bioterrorism/threat.html. Accessed 30 July 2023.
Johns Hopkins Center for Health Security. (2023). Factsheet – Bacillus anthracis. https://centerforhealthsecurity.org/sites/default/files/2023-02/anthrax.pdf. Accessed 30 July 2023.
Github – Manoharam Lab Harvard University. HoloPy. (2021). Hologram processing and light scattering in Python. https://manoharan.seas.harvard.edu/holographic-microscopy, https://github.com/manoharan-lab/holopy. Accessed 30 July 2023.
International Telecommunication Union. (2021). Deep learning software framework evaluation methodology standard (ITU-T F.748.12). https://www.itu.int/rec/T-REC-F.748.12/en. Accessed 30 July 2023.
Go FAIR. (2020). FAIR principles. https://www.go-fair.org/fair-principles/. Accessed 30 July 2023.
Acknowledgments
The project was supported by Horizon 2020 program: HoloZcan (GA: 101021723). (https://datasenselabs.net/; https://datasenselabs.net/horizon2020/).
Special thanks to the microbiology and CBRN experts of the University of Lodz, Poland for their contribution in the air sample collection and sample preparation.The results summarized in this publication are based on the state of the R&D process presented at the Research & Innovation Conference for European Security and Defense (https://rise-sd2023.eu/) 29–31 May 2023, Greece.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Copyright information
© 2025 The Author(s)
About this chapter
Cite this chapter
Pálhalmi, J., Mező, A. (2025). AI-Powered Microscopy Platform for Airborne Biothreat Detection. In: Gkotsis, I., Kavallieros, D., Stoianov, N., Vrochidis, S., Diagourtas, D., Akhgar, B. (eds) Paradigms on Technology Development for Security Practitioners. Security Informatics and Law Enforcement. Springer, Cham. https://doi.org/10.1007/978-3-031-62083-6_10
Download citation
DOI: https://doi.org/10.1007/978-3-031-62083-6_10
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-62082-9
Online ISBN: 978-3-031-62083-6
eBook Packages: Physics and AstronomyPhysics and Astronomy (R0)