The application of AI in medical imaging is developing rapidly, and the achievements in technology and commercialization are emerging. On March 26, China Medical Imaging AI industry university research innovation alliance convened imaging experts, scientific research experts and leading AI medical companies from domestic top three hospitals to jointly draft the white paper on Chinese medical imaging AI, which was officially released.
The survey population of the report is radiation doctors, imaging AI related researchers and enterprise personnel. The public platform survey and targeted questionnaire survey are adopted to understand the current situation and needs of China’s medical imaging AI industry. A total of 2135 hospitals and 5142 doctors were investigated, of which 59% were tertiary hospitals and 38% were secondary hospitals; There were 120 researchers.
The report shows that young and middle-aged doctors, senior doctors and radiology managers generally pay more attention to AI technology. Most hospitals lack AI research institutes or achievement transformation departments, and only 1% of the surveyed hospitals have established AI research departments; 88% of medical imaging AI products are concentrated in pulmonary nodules; The primary resource that doctors can provide in cooperation is images; Most doctors believe that the biggest problem in this field is the lack of industry standards and AI related knowledge.
Liu Shiyuan, chairman of China Medical Imaging AI industry university research alliance, said that compared with the traditional medical model, the development and popularization of AI medical technology is very necessary.
“There are two problems in China’s medical system. One is the lack of medical resources, and the other is the uneven distribution of resources. Medical imaging AI may be an opportunity.” Liu Shiyuan believes that medical imaging AI is a working mode based on cloud platform, which will change the whole medical imaging in the future and bring intelligent, Internet and clinical changes. Of course, once the AI model is mature, it forms a more accurate diagnosis model in developed areas and high-level hospitals. It can also obtain the same diagnosis effect in remote areas. For areas with few grass-roots doctors, it can greatly improve the discovery rate and diagnosis accuracy of diseases.
Artificial intelligence has many applications in the medical field, including the application of natural language recognition and speech recognition in intelligent diagnosis guidance and medical record input, and the application of big data analysis in pharmacy, drug discovery and clinical analysis. At present, more attention is paid to the application of medical images. From the reconstruction of equipment images to the reading of images, the diagnosis of fundus diseases, brain diseases and adjuvant therapy, they have begun to have commercial applications.
The report points out that AI algorithms for medical images have made great breakthroughs and made new progress in auxiliary physiotherapy, universality and open AI computing platform. Taking the open AI computing platform as an example, the hospital oriented medical AI platform is mainly divided into two directions: the image AI platform for medical image data processing; Data AI platform for NLP. On the basis of these two kinds of platforms, we integrate the Internet, cloud computing, AI, big data analysis and other cutting-edge technologies to realize the cloud collaborative sharing of high-quality medical resources and the deep mining application of massive diagnosis and treatment level big data, and customize a series of cloud intelligent solutions for the government, hospitals, scientific research institutions and individuals. For researchers and developers, the corresponding platform provides basic services such as medical AI model modeling, training and open application, so as to promote the development of medical AI industry.
The report also points out that there are some prominent problems in the current medical imaging AI application, such as less data, inconsistent standards and nonstandard labeling. These problems need to be solved as soon as possible.