64% of radiologists agree with the results of AI algorithms

64% of radiologists agree with the results of AI algorithms

IT NewsFacts and forecastsTechnology

Ekaterina Alexandrova | 09.11.2020

64% of radiologists agree with the results of AI algorithms

Over the past six months, the number of radiologists who are active users of medical services based on computer vision has increased. The number of skeptical doctors to
artificial intelligence technology dropped to 36%. These results were reported by Sergei Morozov, Moscow’s chief specialist in radiation and instrumental diagnostics, director of the Center for Diagnostics and

The results of the work of AI services as part of the Moscow experiment on the introduction of computer vision in radiation diagnostics are evaluated by half a thousand radiologists. Moscow implementation experiment
computer vision in radiation diagnostics has been carried out since February 2020. This is a large-scale scientific study of the applicability and quality of artificial intelligence for solving practical problems
health care. The results of the experiment will be announced next year.

New data showed that 64% of radiologists agree with the results of artificial intelligence algorithms. Other users are still dissatisfied with the quality of medical treatment
images – the results, according to their version, are erroneous. Some doctors believe that the occurrence of false positive and false negative results in the processing and analysis of medical images
artificial intelligence is associated with the quality of IT development, and it is necessary to “retrain” algorithms. Sergey Morozov calls the real reasons for the erroneous diagnosis problems from outside
developers: gaps in the quality management system and methodically incorrect selection of datasets for training algorithms.

“Not all professionals start to quickly support new technologies; every IT product adapts to change through the innovation curve. In addition, there are doctors who express concerns about
that the profession of a radiologist is in danger in the future. This is not true. Obviously, in the future, radiologists will work with a large number of complex studies,
as the volumes of diagnostics and types of research are growing. And the share of artificial intelligence will have routine, monotonous tasks, for example, preventive research, “says Sergey

The problem of a high flow of research and, as a result, a staff shortage can be solved by computer vision technology aimed at reducing the time required to describe medical images. IN
Currently, 33% of radiologists are convinced that algorithms can reduce the time required to prepare reports. However, another 32% of radiologists believe the opposite is true: in their opinion, time is
the description has increased due to AI, since not all the functionality of the services is finalized. Nonetheless, 46% of physicians are confident that AI services will help reduce the absence of clinically significant
discrepancies. In the future, the operator of the experiment, the Diagnostic and Telemedicine Center, is considering creating personalized worklists for doctors depending on their specialty, within the framework of
which radiologists will be able to choose for themselves certain algorithms, but such a model must be tested.

According to a study by the Center for Diagnostics and Telemedicine, the average time for processing medical images by services is 10 minutes. Health standards for planned description
X-ray examinations are allocated 24 hours. “The real time of preparation of the conclusion can take less time due to the remote description of studies, concentration of expertise in
Reference Center and process automation, ”said Anton Vladzimirsky, Deputy Director for Research at the Center for Diagnostics and Telemedicine. This is especially true for situations where
the duration of the preparation of the X-ray report can be up to a week, and the records for examinations are not opened until the presence of a doctor on the spot is ensured.

Another objective of the digital project, as Anton Vladzimirsky concluded, is to reduce the burden on doctors to process a large flow of routine examinations with a low percentage of revealing significant
pathologies. According to him, out of three thousand scanned fluorograms, only one case with signs of tuberculosis can be identified. For this reason, the experiment included such routine and massive
studies like fluorography, mammography, lung cancer screening, which can be automated using computer vision technology. Together with this, in the future, some functions
radiologists can be transferred to other specialists. In particular, an X-ray of the lungs with automated marking of pneumothorax will immediately be able to go to an intensive care physician,
who, in accordance with the professional standard, is entitled to a diagnosis.

Artificial intelligence, medicine

Leave a Comment