Microscope-microbiologist
A "smart" microscope has been taught to recognize bacteria in the blood
Ksenia Malysheva, Naked Science
Microbiologists from Beth Israel Medical Center (BIDMC) have created a "smart" microscope that uses artificial intelligence to detect pathogens in the blood. Images obtained using a microscope are processed by a machine learning algorithm.
A convolutional neural network that recognizes images taken with a microscope was trained on an array of 100 thousand fragments of images of 25 thousand samples, pretreated with dye to make bacteria more contrasting, and aged under normal conditions in order to give bacteria the opportunity to multiply. The neural network has learned to recognize bacteria and determine their appearance by external signs. The neural network took into account the shape and organization of bacteria, because by these signs it is easy to recognize the most common pathogens – E. coli sticks, rounded clusters of staphylococci and streptococcal chains. By the end of the training, the accuracy of determining the types of bacteria was 95%.
Then the neural network was tested on 189 samples that were processed without human intervention; in this case, the system correctly identified the type of bacteria in 93% of cases. Further training can turn the system into a reliable and completely autonomous diagnostic tool, the authors of the development believe.
The authors of the development note that laboratory technicians, as a rule, very accurately diagnose bacterial infections by blood tests. It was not necessary to attract AI in order to increase the efficiency of people's work. The fact is that microbiological laboratories are experiencing a great shortage of qualified laboratory assistants, and a new influx of personnel is not expected in the coming years. Artificial intelligence can take the place of missing specialists; this will allow patients not to face the consequences of a personnel crisis, Israeli scientists explain.
Bacterial infections kill thousands of people around the world every year; timely, accurate diagnosis can save their lives. In addition, a trained neural network itself becomes a database on which new specialists can be trained. In addition, it can be used for data analysis and generalization and for scientific work.
The description of the system is published in the Journal of Clinical Microbiology (Smith et al., Automated Interpretation of Blood Culture Gram Stains using a Deep Convolutional Neural Network – VM), a press release on the BIDMC website briefly describes it.
Portal "Eternal youth" http://vechnayamolodost.ru