Research Assistant University of Rochester Essex, Connecticut, United States
Introduction:: Open wound bone fractures can be susceptible to bacterial infection, with Staphylococcus aureus (S. aureus) being the most common cause. S. aureus can invade the Osteocyte Lacuno-Canalicular Network (OLCN) where few treatment methods have proven effective. Since antibiotics usually cannot reach the infection site, amputation of the infected region is the most common outcome. The bacterial diameter of S. aureus is much larger than the bone canaliculi that it infects (1-1.5 microns vs < 1 micron) and S. aureus' only means of creating force is through division. Creating a model of bone infection is crucial to understand how these pathogens infect sub-micron spaces and may aid in understanding how to develop relevant therapies.
To study the progression of S. aureus infection, we use a 400 nm thick 0.5μm porous silicon membrane. The membrane's thinness allows us to approximate two dimensional slices of the OLCN, providing an appropriate model for OLCN invasion. Staphylococcus is seeded over the membrane in a microfluidic device featuring a silicon membrane canalicular array ( μSiM-CA) and bacterial transmission throughout the device is imaged with a confocal microscope.
By preprocessing dataset images using adaptive thresholding, constructing a multi-layer convolutional neural network and training on a large dataset of bacterial images, it is possible to count the amount of bacteria present on the bottom channel with similar accuracy to a human expert. This allows us to produce a quantitative analysis of bottom channel images, providing a precise understanding of how original seeding concentration affects the bacterial transmission level.
Materials and Methods:: GFP expressing S. aureus (USA300) were added to the well of the μSiM-CA. The device was then sealed with a PDMS lid and polymer sealer - preventing contamination while maintaining an aerobic environment to foster cell growth. Bacterial transmission was monitored through the membrane for up to 6 hours. Confocal microscope images were taken using an Andor dragonfly spinning disk confocal microscope. Images were taken at different time points throughout the monitoring period, with a 1 micron step size between images in the z-dimension. Images were processed using Imaris imaging software. Images from the device bottom channel were then exported and preprocessed using adaptive thresholding techniques found in the OpenCV Python Library. Preprocessing will aim to remove image noise generated from the confocal microscope, thus creating a training and testing dataset. Features of Staphylococcus bacteria will be used to produce segmented data from the preprocessed dataset. A portion of the dataset in combination with segmented image data will be used to train a multi-layer convolutional neural network. Performance of the neural network will be assessed using unused images from the generated dataset.
Results, Conclusions, and Discussions:: Successfully using a convolutional neural-network to analyze images of Staphylococcus aureus would allow us to gain a quantitative analysis of how bacterial transmission varies as a function of original seeding concentration in our OLCN model. This would be a particular improvement over our current binary result metric; a qualitative analysis where bottom channel images are checked for presence of bacteria - without discriminating by bacterial count. It has been previously found that convolutional neural-networks have a rough accuracy of at least 70% in classification in similar bacterial datasets. Processing images with such accuracy on a large scale dataset would allow us to achieve a quantitative analysis reinforced by a large dataset, where manual counting would be impossible due to time constraints.
Acknowledgements (Optional): :
References (Optional): : C. Spahn, E. Gómez-de-Mariscal, R. F. Laine, P. M. Pereira, L. von Chamier, M. Conduit, M. G. Pinho, G. Jacquemet, S. Holden, M. Heilemann, and R. Henriques, “DeepBacs for multi-task bacterial image analysis using open-source deep learning approaches,” Communications Biology, vol. 5, no. 1, 2022.
E. A. Hay and R. Parthasarathy, “Performance of convolutional neural networks for identification of bacteria in 3D microscopy datasets,” PLOS Computational Biology, vol. 14, no. 12, 2018.
E. Jelli, T. Ohmura, N. Netter, M. Abt, E. Jiménez‐Siebert, K. Neuhaus, D. K. Rode, C. D. Nadell, and K. Drescher, “Single‐cell segmentation in bacterial biofilms with an optimized deep learning method enables tracking of cell lineages and measurements of growth rates,” Molecular Microbiology, 2023.
E. Jelli, T. Ohmura, N. Netter, M. Abt, E. Jiménez‐Siebert, K. Neuhaus, D. K. Rode, C. D. Nadell, and K. Drescher, “Single‐cell segmentation in bacterial biofilms with an optimized deep learning method enables tracking of cell lineages and measurements of growth rates,” Molecular Microbiology, 2023.
