Launched in Moscow clinics in September 2020, the medical decision support system – a digital assistant based on artificial intelligence – has become even more accurate, said Anastasia Rakova, Deputy Mayor of Moscow for Social Development. According to her, now the coincidence in the choice of a preliminary diagnosis by the therapist and the neural network is 46%.
“The medical decision support system became an integral part of the work of Moscow physicians ten months ago. During this time, the neural network has trained itself for 12 million visits, and an improved version of the service has been launched in all Moscow polyclinics since the beginning of May. It became more accurate by 7.4% and, as a result, more convenient for a doctor's work, which has a positive effect on the quality of services for patients. The development of digital services in healthcare as a whole helps to reduce the likelihood of errors in making a preliminary diagnosis and to increase the detection of diseases at the primary care level, ”said the Deputy Mayor.
The medical decision support system is represented by two modules. The module for making a preliminary diagnosis analyzes the patient's complaints and selects, based on them, one of the three most possible diagnoses. In this case, the doctor can either choose one of the proposed diagnoses, or put his own. The accuracy of the neural network is 68% when choosing from three diagnoses and 46% when choosing from one.
Thanks to the second module, the doctor receives information about which instrumental and laboratory tests need to be prescribed and which specialized specialists need to be consulted to clarify the diagnosis. At his discretion, the doctor can shorten the list or prescribe additional tests.
To create a neural network, the statistics of all diagnoses in city polyclinics for one year were summarized, more than 2 million visits were analyzed. The use of the service makes it possible to reduce the average time of the appointment of diagnostic examinations by ten times.
Read also:
