Artificial intelligence is highly skilled at image recognition and pattern recognition tasks when provided ample training. AI lacks some of the innate vulnerabilities and shortcomings that humans have. Humans can be vulnerable to fatigue, stress, and a multitude of other factors which may limit their effectiveness in work, despite consistently accurate diagnoses under normal conditions.
Microbe identification is a task which can easily absorb the time of a skilled microbiologist, especially in a high-population area, where infection risks are higher and pathogens can be exponentially more prevalent. Microbiology is a field which sees a national average of 9% of lab technologist jobs unfilled.
When dealing with illness, time is of the essence, a major reason why having an automated microbe identification network will enable microbiologists to spend significantly less time sorting through slides unless there is a sample which needs to be retested, changed out, or cannot be directly identified.
How was the network trained?
The convolutional neural network (CNN) was presented with more than 20,000 samples out of more than 100,000 the team prepared. In preparing the sample images, the researchers took images of tests from routine clinical work and cropped the images.
After the neural network was trained on the sample set, the team introduced 189 slides without human intervention. The network was trained to identify images by shape, sorting them into three classifications: Rod-shaped, round clusters, and round chains or pairs.
According to the team:
“The tool becomes a living data repository as we use it.”
How accurate is it?
The network scored exceptionally well on both the control and test sets. It scored 95% accuracy on control images from the over 100,000, and 93% accuracy on the test images. These accuracy levels indicate a promising future in clinical microbiology. Kirby and colleagues had the following suggestion about the network:
“The AI-enhanced platform could be used as fully automated classification system in the future.”
In the test, the neural network was accurate to a minimum of 93% across all three test categories:
How can the network improve the industry?
Having a network which can identify infections in seconds with extremely high accuracy benefits clinics immensely. Being able to send in images for analysis from virtually anywhere benefits the industry as a whole. The highest level of expertise becomes available at local clinics, wherever the internet goes.
Accessibility is vital because both timely identification and delivery of medicine are key to treating bloodstream infections. These infections can kill as hosts up to 40% of the time.
By reducing lab technicians’ time investment in identification the demand for skilled labor in that area can be reduced according to market availability. This is vitally important when considering that nearly 20% of workers in this role will be reaching the age of retirement in the next five years.