Artificial Intelligence Enhanced Microscope Identifies Bacteria with 95% Accuracy

Artificial Intelligence Enhanced Microscope Identifies Bacteria with 95% Accuracy

An artificial intelligence enhanced microscope is being used at Beth Israel Deaconess Medical Center in Boston to speed up the process of identifying bacteria in clinical samples.

Currently, identifying bacteria requires the trained eye of a clinical microbiologist, although they are currently in short supply. Currently, almost one in ten microbiologist positions at hospitals remain unfilled due to a lack of highly trained microbiologists, resulting in delays diagnosing potentially fatal blood infections.

Clinical microbiologists in hospitals spend a considerable amount of their working day on the time-consuming process of examining specimens to determine the type of bacterial infections patients have. That information is then relayed to doctors who prescribe the appropriate antibiotics. Speeding up the process of identifying bacteria means treatment can be started more quickly and patient suffering is reduced. Since as many as 40% of people with bacterial blood infections are killed as a result, rapid identification of bacterial infections could save many lives.

The lack of skilled microbiologists is only likely to get worse in the short term. Figures from the American Society for Clinical Pathology suggest 20% of lab technologists are approaching retirement age and will retire from the profession in the next five years and there are not enough trained microbiologists to take their place.

However, the new artificial-intelligence based approach could help to alleviate the problem, with machines taking on a considerable amount of the workload. The AI system trialed at the Boston hospital and has produced highly promising results.

The artificial intelligence enhanced microscope was developed by MetaSystems and uses a digital camera to capture images of bacteria in samples in high resolution. The system is able to compare the images to images in its database, and based on the shape and distribution of the bacteria, is able to diagnose which types of bacteria are present.

The system has been used to identify bacteria as rod-shaped, round clusters, and round chains and pairs with 95% accuracy with some human assistance sorting the slides, and with 93% accuracy on new samples without human assistance.

The system uses a type of AI called convolutional neural network (CNN). While the system must be trained to identify bacteria, the more it is trained, the more accurate its determinations become. Training the system involved more than 100,000 images, with the system getting better and better, as training progressed.

The artificial intelligence enhanced microscope is unlikely to replace microbiologists in hospitals, but it could perform a considerable amount of the time-consuming work and aid microbiologists making diagnoses.

James Kirby, MD, director of the microbiology lab at Beth Israel Deaconess Medical Center said with further development of the system and more training it could be possible to develop a fully automated bacteria classification system, “conceivably reducing technologist read time from minutes to seconds.”

The trial of the system has been detailed in the paper – Automated Interpretation of Blood Culture Gram Stains using a Deep Convolutional Neural Network – has been published in the Journal of Clinical Microbiology.

Leave a Reply