AI is helping medical care become smarter and more convenient

Author: Fan Xinru Source: IT Times

Among all industries, the medical industry is one of the industries that generates the most data. The amount of data it generates in a year accounts for approximately 30% of the total global data. Among them, smart medical care generates about 59PB of data every year, and biomedical research generates about 40PB of data every year.

At the 2023 China International Medical Equipment Expo and the "AI Edge Computing Empowers Medical Imaging and Assists Primary Medical Innovation and Upgrading" forum, Guo Wei, general manager of Intel China's Internet of Things and Channel Data Center Division, believed that how to make good use of this data , is the core of technology-empowered medical care. "In the medical industry, Intel mainly does three things, namely 'one foundation and two core points'." Guo Wei explained that one foundation is to empower the industry based on artificial intelligence and allow artificial intelligence to be implemented on Intel's chips Run better. The two cores refer to accelerating scientific research innovation and empowering smart medical care.

The blessing of artificial intelligence and other technologies is making medical care more intelligent and convenient.

Guo Wei, General Manager of Intel China Internet of Things and Channel Data Center Division

AI detection rate for certain single diseases can reach 95%

With the application of AI in the medical industry, AI has been widely used in medical imaging, mainly focusing on process transformation, disease diagnosis, and health management and treatment.

Liu Shiyuan, chairman of the Radiology Branch of the Chinese Medical Association and chairman of the China Medical Imaging AI Industry-University-Research Innovation Alliance, announced a set of data at the meeting. The results of a national medical imaging artificial intelligence survey showed that the AI ​​usage rate in large hospitals It has reached 73%. Up to now, there have been more than 55 NMPA registration certificates in my country.

AI is making medical examinations more convenient. For example, in coronary imaging diagnosis and treatment, when AI was not used in the past, it often took about 30 minutes for a patient to go from scanning to image reconstruction to generating a report. But now, because of AI, the reconstruction process has been shortened to about 1 minute. In other words, it now only takes 6 minutes to complete the entire process from coronary artery examination to report generation, which greatly increases the efficiency of the examination. Liu Shiyuan said: "The faster examination speed means that the hospital can complete more coronary examinations in one day than before, and patients can benefit from this."

In the detection of pulmonary nodules, coronary artery CTA imaging, head and neck CTA artificial intelligence model, fracture model, perfusion, etc., which are often referred to as the "Five Diamonds" in the medical community, AI already has a high detection rate. In Liu Shiyuan's view, AI has advantages in some parts that are easily blocked. Because AI has a higher resolution, it is generally easy to miss diagnoses with the naked eye in places that cannot be seen on flat films, but AI can "see" it more clearly and alert doctors of possible lesions. He gave a statistical data from a tertiary hospital. Data show that in the coronary field, the application of AI has helped the hospital's plaque stenosis detection rate increase from more than 60% to about 95%. The detection rate of pulmonary nodules increased from 35% to 70%.

The products provided by Huiyi Huiying Company can help patients complete bone density related tests while taking CT scans. Chai Xiangfei, CEO of Huiyihuiying, said that without using a body membrane, patients only need to scan the chest or abdomen, and AI can automatically calculate bone density. Today, this test can be widely used in osteoporosis screening, including pyramidal morphology analysis, fracture prediction, sarcopenia, obesity, and preoperative planning.

Huiyi Huiying CEO Chai Xiangfei

Although AI is currently mostly used in the diagnosis and treatment of single diseases, Liu Shiyuan believes that with the further development of artificial intelligence imaging, AI medical products will become more and more abundant in the future. These products will help AI diagnosis and treatment develop from a single disease to multiple diseases and multiple tasks, forming an integration of software and hardware based on parts and organs, Internet-based source sharing, and integration of diagnosis and treatment, thereby achieving full-process coverage and forming a good ecosystem. .

"The development of AI is coming at you," Liu Shiyuan said, "just like ocean waves, hitting one after another. Each wave looks the same, but its connotation is different."

AI will not replace doctors in 10 years

When AI becomes more and more involved in medical treatment, does it mean that traditional radiologists will have their job functions replaced?

"In 2016, when we started to build the Medical Imaging Artificial Intelligence Alliance, colleagues would say that what you are doing is being your own gravedigger." A medical industry expert said, "But in the past five or six years, Judging from our experience, at least within ten years, it will not be able to replace imaging doctors."

This medical industry expert believes that although the accuracy of AI has surpassed human doctors in detecting some common diseases. But in actual medical diagnosis, AI still has shortcomings. She said: "Based on the current limited reliable reports that can be seen, if you think about the bright future, AI can reach the level of senior attending doctors."

The reason why she made this judgment is that in her opinion, medical knowledge is completely different from simply looking at images and photos. In actual medical diagnosis, doctors need to combine the patient's medical history, medication history and other multiple information to judge the patient's condition, rather than simply giving a diagnosis based on pictures. Therefore, the report given by AI is only a reference, and doctors still need to rely on their own experience to make comprehensive judgments. "However," she said, "if doctors are only satisfied with simple diagnosis, they will definitely be replaced by AI."

But in Liu Shiyuan’s view, although AI is still in the stage of assisting treatment decision-making, with the development of ChatGPT, the possibility of AI replacing doctors seems to be getting higher and higher. “But I don’t think this is a crisis,” he said. “This is actually an opportunity.”

In his vision, doctors in the future will not just face the cold computer screen, but will need to live a more vivid life, communicate with patients, and solve problems for patients. He said: "The value of doctors lies in interpreting the reports generated by AI together with patients." When the accuracy of AI reports is high enough and can be directly used for disease diagnosis, it means an increase in efficiency for doctors and patients, so doctors Both the patient and the patient have more time to devote to work and life. “Isn’t this very convenient?”

Multimodal AI is the future

How far is there between assisted diagnosis and AI autonomous diagnosis? The answer given by an Intel expert is that there is still a lack of multi-modal AI.

In the opinion of this expert, one reason why imaging AI can usually only be used as an assistant is that imaging diagnosis itself is not based solely on imaging data. He said: "Whether you are a radiologist or a clinician, when formulating a diagnostic plan, you need to combine the patient's basic information, demographic information, historical case information, and even genetic data, and make a comprehensive assessment to arrive at a diagnostic result."

This means that relying solely on imaging AI cannot solve the diagnosis problem. It must be combined with electronic medical record data, genetic data and other data sources for multi-modal analysis. "This is also a trend in future clinical applications." He said.

In the medical industry, although multi-modal analysis has been carried out for many years, the current research on multi-modal AI is still in the research stage and has relatively few applications. This is because multimodal analysis usually involves multiple data such as images and cases. In the medical community, data acquisition and sharing remains a worldwide problem.

Compared with large models such as ChatGPT, the data feeding volume of AI products in the medical industry is often around 1,000 cases, which is far less than the data feeding volume of GPT. This poses a challenge to the generation of medical multi-modal AI. When it comes to data annotation, the establishment of standards in the medical industry is far more difficult than in other industries. Even among expert-level doctors and professors, it is difficult to reach an agreement on the same disease. All these pose challenges to the generation of medical multi-modal AI.

At the theoretical level, the process of multi-modal AI diagnosis is to input multi-modal data. After AI fuses the data, it uses the model to generate a diagnosis. “Because it combines information from different dimensions, theoretically, its results should be more accurate,” the Intel expert said. “But the problem is that when the model becomes more complex, it may become super large.” This will not only improve the model The difficulty of research and development will also increase the difficulty of use by users. Because the understandability and interpretability of multimodal models will be worse.

The above is the detailed content of AI is helping medical care become smarter and more convenient. For more information, please follow other related articles on the PHP Chinese website!