Doctors use machine learning to fight deadly infections

Modern technologies allow making computer systems more and more “smart”. In particular, machine learning is used by Facebook, Google, Amazon — for example, in order to convert speech into text, recognize faces and give personalized advertising. These technologies also help doctors of various specialties. In the USA, machine learning and a weak form of AI are planned to be used to combat the dangerous bacterium Clostridium difficile . Wikipedia says that the bacterium is the causative agent of pseudomembranous colitis, a serious infectious disease of the intestine that can lead to the death of a person with a weakened body.

Hospital patients fall into the category of people with a weakened body, who are almost unable to fight off pathogenic viruses and bacteria. In the United States alone, 453,000 cases of colitis are recorded annually, with 29,000 cases leading to the death of the patient. In any case, such data was presented in the report for 2015 . Conventional medical methods do not help stop the infection - washing hands, floors, antiseptics, etc. not too effective.

For this reason, the doctors decided to use other methods that are directly related to modern technologies, which were discussed above. A team of the University of Michigan has developed a method to predict the likelihood of colitis in different patients. For this purpose, medical data from a person’s card is used. So far, the technology is only being tested, but it has already shown excellent results.

A few weeks ago, 374,000 cases of hospitalization at the Massachusetts Hospital were analyzed using computer-generated techniques. Scientists want to find a connection between the occurrence of the disease and the factors that led to the disease.

During the analysis, the machine checks about 4,000 different factors, including the placement of the patient's bed, his visits to the doctor, the placement of the beds of other patients and all other data. Usually patients who are in the hospital are not located in one place, but move around the hospital. And if somewhere there is a source of infection, then it works. The main thing is to identify this source in time.

Another example of the use of machine algorithms in medicine is the analysis of the retina for the detection of diabetic retinopathy. Retinopathy is a lesion of the retinal vessels, leading to impaired blood supply to the retina, its degeneration, optic nerve atrophy and blindness. This is a common disease for diabetics that needs to be diagnosed on time so that people do not lose sight as a result of progressive retinopathy. It is quite common - in 2011 about 126 million cases were recorded , by 2030 their number could increase by 51%.

Experts have created a neural network, which was taught by the example of 128 thousand images of the eyes of patients in India and the USA. Before acquaintance with them, the same pictures were shown to doctors who diagnosed retinopathy at different stages of development. Then all the data was downloaded to the computer. After the neural network was trained, it was checked on another data set, and in this case its results were higher than those of the doctors.

Now the team of specialists who developed this tool is planning to implement it in India, where there are a large number of patients with this diagnosis. And in the United States or other countries, this is also a problem. The technology is currently undergoing clinical trials in a number of hospitals in India.

Last year, the US regulator approved the introduction in hospitals of a special apparatus for analyzing magnetic resonance imaging. The device analyzes the image in just 30 seconds, while the doctor spends about 45 minutes on the same work. “This is an automation of routine processes that absorb a large amount of specialist time in a normal situation,” said Carla Leibovitz, a spokeswoman for Arterys, who developed this system.

In general, machine learning and AI are becoming increasingly widespread in medicine. Basically, we are talking about the automation of any routine processes that are important for diagnosing a disease, but which take a large amount of doctor time. And the more time a doctor spends on such work, the more tired. As a result, the same X-ray images are analyzed by a person already with much less attention after work for several hours than at the very beginning of work. Accordingly, the accuracy of diagnosis drops.