"Urgently recruit sonographers, provide accommodation, and receive a talent reward of up to 300,000 yuan."
Recently, a hospital in Longhua District, Shenzhen put out such a recruitment advertisement on its official account. Posts similar to urgently recruiting ultrasound doctors are also common on the Internet, which reflects the current shortage of ultrasound talents.
According to public data, there is a shortage of at least 150,000 ultrasound talents in my country.
The workload of medical image analysis is heavy and cumbersome, which consumes doctors' energy extremely.
In recent years, artificial intelligence (AI) is being applied in the medical imaging industry, with the purpose of helping "liberate" doctors' hands, but the specific application in the field of ultrasound is not satisfactory. Currently, it has obtained the third category registration of medical devices from the State Food and Drug Administration. Among the certified AI medical imaging products, there are still no ultrasound-related products, and the approved products are mainly concentrated in the fields of X-ray and CT.
Ultrasound is the most widely used in medical imaging diagnosis.
In the development of AI medical care in my country, medical imaging has become the most popular application direction, but why does it progress slowly in the subdivision field such as ultrasound?
Recently, six departments including the Ministry of Science and Technology issued the "Guiding Opinions on Accelerating Scenario Innovation and Promoting High-quality Economic Development with High-level Application of Artificial Intelligence", proposing that efforts should be made to solve major application and industrialization problems of artificial intelligence, and comprehensively improve the quality and quality of artificial intelligence development. Among them, the medical field should actively explore scenarios such as medical imaging intelligent auxiliary diagnosis, clinical diagnosis and treatment auxiliary decision support, medical robots, Internet hospitals, intelligent medical equipment management, intelligent hospitals, and intelligent public health services.
Challenges in the development of AI ultrasound
Challenges in the development of AI ultrasound
At present, the medical imaging equipment with wide clinical application includes four types: X-ray, CT, magnetic resonance, and ultrasound. Among them, ultrasound is widely used in clinical practice based on the advantages of safety, non-invasiveness, real-time, economy, and portability.
Ultrasound medical imaging equipment is also a commonly used clinical diagnostic instrument in hospitals, imaging centers and other medical institutions. The application field has expanded from early abdominal and obstetrics and gynecology diagnosis to clinical diagnosis in cardiovascular, neurological, musculoskeletal and other fields, and gradually penetrated into Ultrasound-guided intervention and other non-diagnostic fields.
There are more than 200,000 ultrasound doctors in China. Correspondingly, the number of ultrasound examinations per year reaches about 2 billion, far exceeding the number of CT examinations of 200 million per year.
Ultrasound examination market demand is large, but it is not easy to make an appointment for examination. The reason is the lack of ultrasound doctor resources, and it often takes 3 to 5 years to train a qualified ultrasound doctor. In the field of prenatal fetal screening, it may even take 5 to 8 years. year time.
Ultrasound examinations are mainly performed manually, rely heavily on the doctor's experience, and have high technical requirements for the doctor.
In February 2020, the US FDA approved the artificial intelligence ultrasound imaging assistance system developed by Capture Health, bringing a breakthrough to the application of ultrasound imaging.
In recent years, some Chinese companies are also deploying the field of AI ultrasound, trying to use AI to assist ultrasound diagnosis.
However, as of now, compared to other AI medical imaging tracks, the competition in the AI ultrasound track is far less intense than expected.
"In the development of the AI imaging industry, AI ultrasound started late, developed slowly, and was difficult to commercialize. The pain point that restricts the development of AI ultrasound is that it does not match clinical needs. One of them is the difficulty in realizing real-time diagnostic functions." Zhongshan Xie Hongning, professor and chief physician of the Obstetrics and Gynecology Ultrasound Department of the First Affiliated Hospital of the University, told Yicai.com that the collection and diagnosis of radiology images such as CT, MRI and X-ray are separate. Completed by radiologists, the latter can be diagnosed by judging static images, but the difficulty of ultrasound diagnosis is that image acquisition and reading need to be completed at the same time. Perform real-time diagnostics.
It is very difficult to develop AI ultrasound to assist ultrasound doctors in real-time diagnosis.
"For example, in the field of breast cancer screening, most of the AI ultrasound products require doctors to scan and find the image of the tumor, capture it into a static image, and then use the AI system to determine whether it is benign or malignant. When doctors can't see the lump, there is a risk of missed diagnosis." Li Anhua, vice president of the Chinese Society of Ultrasound Medical Engineering, president and professor of the Guangdong Ultrasound Medical Engineering Society, told Yicai.com.
For ultrasound doctors, if they still need to capture static images and then hand it over to the AI system to judge, not only will it not improve work efficiency, but it will increase the workload, and the ultrasound diagnosis work itself is cumbersome enough.
In addition to the lack of real-time diagnostic capabilities, the current AI ultrasound also has shortcomings in the ability to recognize three-dimensional structures.
"In the process of scanning images, ultrasound doctors need to establish three-dimensional thinking ability, and need to establish three-dimensional structures in the brain to make judgments, but most of the current AI ultrasound systems have not yet achieved the ability to recognize three-dimensional structures." Xie Hongning said.
In Li Anhua's view, the reason why there is a big disconnect between AI ultrasound and clinical applications is that on the one hand, it is related to the scarcity of compound talents. The developers of AI ultrasound products are mainly engineers, who do not have clinical practice experience and are unable to perform well. Understand the real needs of the clinic; on the other hand, it is also limited by the algorithm framework.
"The algorithm framework is strongly related to the accuracy and real-time performance of AI analysis products, and now almost all AI companies in China use open source algorithms, and the algorithm performance of each company still depends on the quality of the recompilation of open source algorithms. How, and recompiling work, is a big challenge in itself."
How to break the game
How to break the game
Although there are still some difficulties in the development of AI ultrasound, it does not mean that there is no clinical demand.
"We are faced with the 'three mountains' of medicine, teaching and research. In addition to undertaking medical imaging diagnosis, we also have tasks of teaching and scientific research. We hope that AI can be empowered to improve work efficiency." Xie Hongning said.
Li Anhua said that different ultrasound doctors have different scanning techniques, and the quality of collected images will also vary. An excellent AI product, in addition to assisting doctors in auxiliary diagnosis, can also assist in the quality control of ultrasound images. , to establish a unified standard for image acquisition.
Specifically at the hospital side, there are two major application scenarios for AI ultrasound. First, in hospital quality control, AI assistance can play a supervisory role. Taking prenatal ultrasound as an example, according to national standards, 30 There are multiple standard sections, and AI ultrasound can judge whether these sections are standard; second, it can solve the problem of shortage of high-quality resources at the grassroots level and help high-qualified doctors train young doctors.
"Currently, the 'siphon effect' of large hospitals still exists, and many ultrasound doctors trained in townships are easily transferred from higher-level hospitals, making it difficult for towns to retain talents. It's important," Li Anhua said.
At present, there are still companies trying to make breakthroughs in the field of AI ultrasound.
For example, in the past July, at the 6th Academic Conference on Prenatal Diagnostics of Guangdong Medical Association, Guangzhou Aiyunji Information Technology Co., Ltd., including the First Affiliated Hospital of Sun Yat-Sen University, joined a number of national top three medical institutions and South China. The School of Computer Science and Technology, University of Science and Technology, jointly developed the ultrasonic AI intelligent system "Love Pregnancy Zhisheng" which took 5 years to jointly develop.
The system has made a breakthrough in achieving real-time synchronization of analytical results.
"It's like preparing a navigation assistant for the ultrasound doctor, which can remind the ultrasound doctor to pay attention to the standardization of the inspection process and abnormal situations through voice, text, images, etc. throughout the process. Regarding the positioning of this product, in terms of the senior doctor group, I hope It can help them reduce some repetitive work and free up time to do creative and decision-making work; for young doctors, I also hope to play a role in quality control training and help them reduce the probability of missed diagnosis and misdiagnosis." Guangzhou Love Pregnancy Record Founder and CEO, Dr. Wang Nan said to the first financial reporter.
Wang Nan also said that it is indeed difficult to realize the function of real-time synchronous analysis of AI ultrasound in the development process.
"At present, the talents who can develop algorithms are mainly concentrated overseas, and these algorithms are mostly used for the recognition of natural images. If they are to be applied to the field of ultrasound, a lot of compilation and optimization work must be done on the underlying algorithms to adapt them to the application scenarios. properties, and after compiling, a lot of testing is required to finally form your own algorithm, which requires a lot of time, energy, resources, etc.”
The currently approved medical device registration certificate for the above product belongs to the second category and can be sold in hospitals, but to achieve more difficult functions, the third category registration certificate is still required.
According to the first financial reporter, in August this year, Aicongzhisheng products also completed the first enrollment of the third class medical device clinical trial of the State Food and Drug Administration in the top three hospitals in Henan and Anhui.
Li Anhua said that at present, in order to truly realize the commercialization of ultrasonic AI, it is still necessary to solve the two major problems of how to improve the product maturity and who will pay.
"We are also exploring the development of services and products with different positioning to adapt to different application scenarios and demands." Wang Nan told the First Financial Reporter.
Looking at the current development of the AI medical imaging industry, although it is still in the period of value practice, the prospects are still optimistic by the market.
A recent report released by the Head Leopard Research Institute shows that AI medical imaging is mainly used in the medical and health market and large health market scenarios. 137.6 billion yuan in 2025, the compound growth rate from 2021 to 2025 is 102.4%.
Zhang Wei, executive vice president and secretary general of Guangdong Artificial Intelligence Industry Association, said that under the joint promotion of national policies and market demand, more excellent companies and products are expected to emerge on the popular track of image recognition-assisted diagnosis.