Medical imaging is involved in all processes of clinical practice.Approximately 70%of diagnostic information originates from radiologic images,which also account for 90%of the digital data volume of a hospital.However...Medical imaging is involved in all processes of clinical practice.Approximately 70%of diagnostic information originates from radiologic images,which also account for 90%of the digital data volume of a hospital.However,the number of physicians has increased only modestly compared to the rapid growth in the number of medical images.In China,data from medical imaging increases by 30%every year,whereas the number of radiologists increases by only 4%annually.Artificial intelligence(AI),which is representative technology of the fourth industrial revolution,may alleviate the increasing pressure and job burnout,and further improve the diagnostic efficiency of radiology services[1].Despite the urgent and realistic demand for AI technology,many challenges remain in the development and translation of AI products.The rate of the scientific translation of AI research into clinical applications is extremely low.Furthermore,AI models that are applied in clinical settings exhibit unreliable performance and are often impractical[2].Therefore,radiologists may not have access to suitable medical imaging AI models to solve specific clinical problems.This paper analyzes and discusses this problem according to two aspects:the data sources and the AI algorithm(Fig.1).展开更多
基金This work was supported by the Scientific and Technological Innovation 2030—a Major Project of New Generation Artificial Intelligence(2020AAA0109503)the Beijing Science and Technology Planning Project(Z201100005620008 and Z211100003521007)+1 种基金the National Natural Science Foundation(82171934)the National Key Research and Development Program of China(2021ZD0111105).
文摘Medical imaging is involved in all processes of clinical practice.Approximately 70%of diagnostic information originates from radiologic images,which also account for 90%of the digital data volume of a hospital.However,the number of physicians has increased only modestly compared to the rapid growth in the number of medical images.In China,data from medical imaging increases by 30%every year,whereas the number of radiologists increases by only 4%annually.Artificial intelligence(AI),which is representative technology of the fourth industrial revolution,may alleviate the increasing pressure and job burnout,and further improve the diagnostic efficiency of radiology services[1].Despite the urgent and realistic demand for AI technology,many challenges remain in the development and translation of AI products.The rate of the scientific translation of AI research into clinical applications is extremely low.Furthermore,AI models that are applied in clinical settings exhibit unreliable performance and are often impractical[2].Therefore,radiologists may not have access to suitable medical imaging AI models to solve specific clinical problems.This paper analyzes and discusses this problem according to two aspects:the data sources and the AI algorithm(Fig.1).
