Feasibility of using artificial intelligence in radiology (first year of Moscow experiment on computer vision)

Abstract.

The rationale for use of artificial intelligence (AI) in radiology departments to analyze medical images in real-life clinical practice was studied in a multicenter prospective trial. This was a part of the “Experiment on the use of innovative technologies in the field of computer vision for the analysis of medical images and further use in the healthcare system of Moscow” taking place in 2020. The trial included 18 different AI systems and 538 participating radiologists, all working within Unified Radiological Information Service. We evaluated applicability of AI systems, demand from radiologists, the quality of AI implementation, radiologists adaptability and AI impact on the overall radiologists productivity.

The final analysis included 1 762 949 AI processing results and 15 028 feedbacks from radiologists.

Commitment of radiologists to use AI systems was 22.4%. Also 65% of the tested AI systems didn’t increase maximal timeline set for the image analysis. AI implementation for analyzing prophylactic mammography images accelerated delivery of the results in outpatient and inpatient setting by 15.0% (p=0.03) and 50.0% (p=0.05) respectively. Lung CT and low-dose CT image analysis (searching for potential lung cancer) took radiologists longer to perform by 42.0% of their standard time (p=0.04) when using AI systems. Such contradictory results of AI implementation in different radiology sub-specialties need to be further analyzed.

Overall the study results suggest time-saving rationale for using AI systems in radiology departments, including emergency settings. The output of AI image analysis should be verified by radiologist.