Rapid identification of seven bacterial species using microfluidics, time-lapse phase-contrast microscopy, and deep learning
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
For effective treatment of bacterial infections, it is essential to identify the species causing the infection as early as possible. Currently, the main techniques for determining the species require hours of overnight culturing of a bacterial sample and a larger quantity of cells to function effectively. This study uses one-hour phase-contrast time-lapses of bacterial growth collected from a microfluidic chip, also known as a ”mother machine”. Then, this data is used to train deep artificial neural networks to identify the species. Both video and image classification models of the Vision Transformer (ViT) and Convolutional Neural Network (CNN) families were evaluated in this study. We have previously demonstrated this approach to classify four different species, which is now extended to seven species: Pseudomonas aeruginosa , Escherichia coli , Klebsiella pneumoniae , Acinetobacter baumannii , Enterococcus faecalis , Proteus mirabilis , and Staphylococcus aureus . The models are then trained and evaluated using subsampled images, simulating using lower-resolution microscopy in a potential clinical setting. The method can be used simultaneously with phenotypic antibiotic susceptibility testing (AST) in the microfluidic chip. The best models attained an average precision of 93.5% and a recall of 94.7% (0.997 AUC) on a trap basis in a separate, unseen experiment with mixed species within 70 minutes. However, in a real-world scenario, one can assume many traps will contain the actual species causing the infection, improving the reliability. The experiments suggest spatiotemporal features can be learned from video data of bacterial cell divisions, and both textures and morphology contribute to the final performance of the models. Rapidly identifying responsible species causing acute infection and simultaneously performing drug sensitivity in a much shorter time than today — reduced from days to hours or minutes — could lead to a paradigm shift in how initial treatments for severe bacterial infections are chosen. These developments could contribute to the fight against antibiotic resistance, improve patient outcomes, and ultimately save lives.
Author Summary
Acute bacterial infections are initially treated by administering many different antibiotics (broad-spectrum) to a patient. Specimens taken from the patient will then be sent to a microbiology laboratory to be cultivated on agar plates, after which the species and the resistance profile to different antibiotics can be obtained. Depending on this information, the patient’s treatment will be adjusted, but the process is work-intensive and typically takes more than 24 hours. However, prompt treatment with suitable antibiotics is critical for the patient’s survival, especially for multi-resistant strains. This study demonstrates a method to speed up this species identification. It utilizes a microfluidic chip, also known as a ”mother machine”, to film seven different bacterial species for one hour using phase-contrast microscopy. Artificial neural network models are then trained to identify the bacterial species in an unseen experiment, only using microscopy video data of bacteria reproducing inside traps of the microfluidic chip. The best model achieved a precision of 93.% and a recall of 94.7%. Furthermore, a large part of the performance was retained when the network was trained to recognize bacteria at low resolution, demonstrating the potential to use the method in a clinical setting using lower-magnification microscopy. The technique opens the door for more effective and targeted treatment of acute bacterial infections and could ultimately save lives.
