Focusing on digital healthcare丨Artificial intelligence large models are widely used, can the "AI + Healthcare" process accelerate breakthroughs?

21st Century Business Herald reporter Tang Weike reported from Guangzhou

With the advent of ChatGPT, the development of big data models has entered a new stage, and AI has penetrated into various fields.

In the medical field, the application of AI can promote the high-quality development of medical undertakings, and the medical field also provides broad development space and commercial value for AI.

Li Xinsheng, Vice President of R&D of Mindray Medical Group, said in an interview with the 21st Century Business Herald reporter, "Medical treatment should be the best field where AI can be implemented, because AI is just needed in the medical field, and AI may be the icing on the cake in other fields, but it is a blessing in the medical field." We know that doctors spend a lot of time on case records, paperwork, and don't really pay attention to patients. With AI, doctors can be freed from heavy paperwork and equipment operation and pay more attention to patients, and we can undertake daily records and documents through AI technology, so that doctors can spend less time to output, see cases with high quality, and meet the needs of hospitals. ”

With the penetration of artificial intelligence big models into thousands of industries, for developers, industrial practice has become an important standard for measuring the value of models, and due to the professionalism and complexity of the medical industry, the development of medical artificial intelligence faces three major challenges: data, algorithms, and computing power.

AI development requires a large amount of high-quality data, and most of the domestic medical data is stored in medical institutions at all levels, the business system is relatively independent, the data is difficult to share, there is an obvious "data island" phenomenon, and the real scene dataset for training is limited.

AI still needs to solve the data challenge

Even if a large amount of medical data is obtained, how to process, statistically and analyze discrete massive medical professional data, and effectively integrate through the model, especially the rigor of the medical industry, requires higher accuracy of the model, which puts forward higher requirements for algorithms and computing power.

Li Xinsheng said, "With the development of IT technology, the problem of data islands has been slowly solved, and hospitals are connected from daily care, electronic medical records and equipment data, without much difficulty. The difficulty lies in the cleaning of data, data mining, because the collection of a large amount of data also has a mess of fake data interference, if these data can not be cleaned, effectively eliminate interference, can not extract accurate results, no matter how good the model is useless, so this is the biggest technical obstacle. ”

The healthcare industry has long been regarded as one of the best scenarios for AI applications. On the one hand, the application of AI medical treatment can reduce the workload of doctors and improve work efficiency. On the other hand, it can also improve the diagnosis and treatment level of grassroots institutions and change the imbalance in the distribution of medical resources.

However, the landing speed of the last round of artificial intelligence in hospitals is not fast, and more applications are concentrated in the field of imaging, but they also face the challenge of low commercial conversion rate.

The developed medical artificial intelligence achievements have the dilemma of large-scale landing, and the production, supply and sales of data and algorithm models lack the support of industrial chain resources, even in the field of medical examination where artificial intelligence penetrates earlier, it has become an important factor restricting the further development and landing of medical artificial intelligence. By building an artificial intelligence development platform, medical data can be integrated to meet the needs of individual and enterprise developers for training, development, application and sharing.

Through the sharing of computing power, algorithms, data, and models, the AI development platform is built to provide services for startups, medical research institutions, individual developers, industry experts and other users. The open platform can meet the needs of the whole process of development, including data processing, development training, model management, and online deployment, from data training to artificial intelligence application deployment.

By bringing together the disease diagnosis data sets processed by medical experts, the platform can realize the safe sharing of sample resources, high-quality medical examination data and case labeling data, and reduce the development threshold. Through stable, reliable, sustainable and innovative cloud service construction, it can meet the training of general models for multi-party developers without data leaving the domain.

Building an artificial intelligence open innovation platform can drive the integration of data, technology, and industrial chain resources, export the core R&D capabilities and service capabilities of medical examination artificial intelligence, and promote the formation of artificial intelligence industry clusters in the field of medical examination with massive medical examination data services and open AI technology services. At the same time, it can support the development of small and micro medical examination AI enterprises in the industry by pooling industrial resources, promoting industry exchanges, accelerating the research and development process, and helping the application landing.

In addition, practitioners need to rethink how to define the role of AI-enabled healthcare.

Artificial intelligence is considered to be a rigid need in the medical field, which can improve the efficiency of doctors and meet the needs of more patients, facing a common global problem: the ratio of doctors to patients is seriously insufficient.

However, unlike the general-purpose big model, the medical big model focuses on serious and cautious medical scenarios, and naturally has a lower tolerance for errors. This also requires that the accuracy and safety of medical large models must be higher.

"The positioning of 'Dr. Watson' is too radical, making decisions on behalf of doctors, which is the main reason for failure." Li Xinsheng said that for a long time, the positioning of artificial intelligence and large models will be the right assistant of doctors.

"At present, many hospitals have actually applied artificial intelligence assistance, and the emergence of large models brings barrier-free interaction of human-machine natural language, which can mobilize a variety of capabilities to solve problems in multiple scenarios, and it is expected that in the next one or two years, there will be more and more AI medical applications based on large models." In the future, not only in hospitals, but also AI family doctors can help you make preliminary diagnoses, recommend specialties, analyze test reports, etc., and the imagination space is infinite. Li Yinghua, vice president of Golden Mile Medical Group, said.

Explore the new development of "AI + Healthcare"

With the development of large model expansion and application, medical AI technology continues to innovate. In the field of imaging, AI provides new technical support for medical detection. Li Xinsheng said that an innovative point is high-quality imaging technology, the picture after imaging, and the picture under the microscope is highly restored, doctors do not need to do re-examination from the microscope, if the abnormal situation with the electronic pictures taken by the enterprise can be re-examined. The second is innovative flight scanning technology, which imprints a blood cell tomography, and then quickly completes the imaging and statistics of a large area of platelets on the picture, and looks at the coagulation reaction through platelets. The third innovation is to improve the recognition rate of blood cell imaging in images with the help of AI algorithms based on machine learning.

With the more extensive application of major data AI in the medical field, the integration and multi-functional role of large models are more prominent. Zheng Yefeng, a distinguished scientist at Tencent and head of Tencent Tianyan Lab, told 21st Century Business Herald, "Before the big model set off, we launched many services in the form of intelligent dialogue, including intelligent registration, pre-consultation, and Q&A assistant. Previously, we launched different dedicated models to serve different tasks, and after the large model came out, we may be able to use a general model to serve all applications, and one model can correspond to multiple tasks, which is an important improvement for us. ”

At present, a number of companies are trying to explore. In 2020, the Department of Science and Technology of Guangdong Province released the third batch of "Guangdong Province New Generation Artificial Intelligence Open Innovation Platform" list, and Jinyu Medical undertook to build the "Guangdong New Generation Artificial Intelligence Open Innovation Platform for Clinical Laboratory and Pathological Diagnosis", which was officially launched on September 12, 2023.

The improvement of AI diagnosis capabilities can provide more help to the medical industry. Li Xinsheng said, "Last year, we held the 'Boundless Walker' academic event, with about 5,000 doctors participating offline and more than 600,000 doctors participating online, comparing AI film reading and manual film reading. Through a large number of human-computer comparison experiments, it is initially proved that the accuracy of machine reading is higher than that of manual, the accuracy of machine reading has reached more than 95%, and the accuracy rate of young doctors is about 80%, AI technology has significantly improved the accuracy of reading, while greatly reducing the time of reading, it took 25 minutes to half an hour to read a film, and the film was read in less than half a minute after using artificial intelligence tools. "AI film reading can speed up and increase the efficiency of medical business.

With the continuous development of big data, artificial intelligence will expand to more applied medical fields, such as intensive care department, emergency department, anesthesiology, etc. "These departments use large models and multimodal data fusion analysis to help doctors find patients as early as possible, treat patients in time, and make reasonable simplifications and adjustments." With this model, we can greatly shorten the learning time of doctors, so that young doctors can achieve the same level of treatment as senior doctors after technical support, which is our goal. Li Xinsheng said.

For more information, please download 21 Finance APP