Clinical application is the only standard for testing medical AI technology
There are many scenes in the hospital, and it is impossible for companies to develop AI products for each scene. Therefore, companies must make trade-offs and follow certain principles when choosing a direction for development – ​​companies must develop AI products from clinical applications.
"We are not likely to develop products for a single disease in a certain organ or system. Because it is not a clinically realistic scenario, it is hard to imagine that AI applications will only "know" certain diseases. The AI ​​needs a certain breadth of disease awareness, which is the technology and products at the clinical application level." Song Jie said.
Clinical pain points and real application scenarios are the goals and source of medical AI research and development.
Taking the CT field as an example, when everyone is busy identifying specific diseases such as “lung nodulesâ€, Hex's heterogeneity sees a more core need in the field: First, different levels of hospitals have medical skills. Different from experience, CT image quality is very different. Secondly, how to identify the lesion area and better reconstruct the image of the lesion area; from these two perspectives, we solve the clinical practical problem. If we are always entangled in the identification of a certain disease, it may deviate from the clinical needs. After all, there are many kinds of lesions in the same part and organs, and only the problem can be solved fundamentally and broadly. Will really accept.
Histoplasma heterogeneity places great emphasis on the real needs of the clinic. "There are too many 'story's of AI in recent years. Now companies with the ability to take out the 'real things' for everyone to see. Heisto isomer can not only come up with different areas of AI application technology, but also is willing to base These technologies develop innovative AI+ medical devices. After all, this is what clinical applications can quickly apply."
"It's the key to taking out the real thing, there is gold under the AI ​​bubble."
As one of the earliest AI R&D companies in the field of digestion, Hirsch's research and development in the field of digestive endoscopy AI covers the entire digestive tract, including tumors, polyps, atrophic lesions, ulcer lesions, erosive lesions, vascular lesions, etc. More than 100 diseases in multiple categories.
In the coming months, Hirsch will introduce endoscopic AI early cancer recognition technology for a variety of different light sources; the areas involved include the upper and lower digestive tract areas reached by conventional digestive endoscopy and the small intestine area exhibited by capsule endoscopy. Related application technologies have been presented in a variety of medical devices (digestive endoscopic AI real-time image determiner, capsule endoscopic image AI analysis and judgment instrument, etc.) and cloud service products, and form a stereoscopic AI product solution in this field.
AI's continuous research and development requires four foundations
After clarifying the direction of AI development, we also need to understand the needs of AI's continuous research and development, or what core competitiveness is required for AI to achieve continuous research and development. Song Jie believes that the following four factors are essential.
1. Medical gene
Medical artificial intelligence is a high-tech field, and it is difficult for AI scientists or clinical experts to do medical artificial intelligence. Successful companies need their managers to have a deep understanding of the industry, such as business models and product needs in the medical segment.
Without understanding these details, it is difficult for companies to design products that meet the needs of doctors. Without understanding the entry modes of medical devices and digital systems, it is difficult to open up markets for products. These need to be accumulated and summarized, and they are also important for the medical gene of the enterprise.
2, artificial intelligence technology
At the first time, AI application technology research and development seems to have a very low threshold. Some data plus cloud computing power can be “started†by using open source algorithms. However, if faced with massive data, it is necessary to develop product-level technologies with real clinical application value. Must have true AI technology capabilities, including: super-computing capabilities, low-level development technology, and application-side AI chip development capabilities, these are hard indicators.
Song Jie believes that an independent super-computing center should be a necessary condition for artificial intelligence enterprises: on the one hand, it can provide powerful computing power, shorten the development cycle, and on the other hand, it can fully guarantee data security.
Take the Hee's heterogeneous super-computing center "Shenong 1 (I)" as an example. This is equipped with 64 NVIDIA's latest TeslaV100 GPUs, with a computing power of more than 90%. The calculation time of the day was shortened to 52 minutes. This is undoubtedly a great improvement for models that require dozens of calculations.
Of course, similar super-computing centers are not simply products that can be bought for money, but rather require companies to build their own technologies.
3. Deep cooperation hospital
Regular business and hospital cooperation often involves risks such as data security, legal aspects, and intellectual property division. Especially in the aspect of intellectual property rights, any previous neglect or legal risk. They may be caught in the throat of the company by others in the future.
Therefore, enterprises should never expect to make valuable products with only a little so-called "data" without the deep participation of high-level medical institutions.
Today, Xi's heterogeneous and in-depth cooperation with many domestic first-class hospitals, taking Huaxi Hospital as an example, the two sides have carried out in-depth cooperation in six areas.
The basic requirements for Hex's heterogeneous cooperation are simple: clarify the legitimacy of data sources and determine the ownership of future AI technologies. This effective insurance mechanism is the basic guarantee for the productization of enterprise R&D results.
4, artificial intelligence productization
If the artificial intelligence fever was a breakthrough in technology last year, then the core word of this year is "landing." Now is the late autumn season, the new year is coming, and the AI ​​will be “commercializedâ€.
AI companies want to be unique, and the ability to hardware is essential. In the modern commercialization mode, AI is difficult to follow the old path of information products, and it is a question worthy of thinking to follow the hardware into the hospital; just like the heterogeneous products of Xi's, Mostly embodied in the form of medical equipment.
Today, there are six products in Hee's heterogeneous that have entered the approval process for three types of medical devices. The era of storytelling of AI medical products will become a past tense, and a new era is coming.
What does AI bring to medicine?
For medicine, is AI a replication ability or a transcendence? This is a topic that many people are confused about.
“In the beginning, I thought that AI can transcend doctors' perceptions of disease. However, during the research of AI, I found it too early to talk about transcendence. At least in the next few years, AI needs to be based on human diseases. Know the ability to copy doctors. But one day, AI can find something unknown to humans." Song Jie said in his speech.
Today, the main function of AI is to use powerful computing power to solve the common extraction that cannot be achieved by conventional methods, and to find out the correlation between disease and phenotype. However, the current AI is far from the intelligence of the Turing experiment.
Song Jie believes that the development of AI now needs to be based on the ability to replicate, and to achieve cognitive transcendence after integration. However, "transcendence" must be done within the scope of human verification. At the time of the AI ​​outbreak, researchers will put a lot of effort into verification.
Taking digestive endoscopy as an example, the work of AI is to analyze the images of the patient's digestive system, and try to find commonalities from the massive images, classify them into rules, and then use machines to enforce these rules and use them in the doctors. The use of verification rules in endoscopy helps doctors analyze images in real time. In this process, the AI ​​replicates the doctor's ability, improves the efficiency of the examination, and reduces the risk of missed diagnosis, but the AI ​​cannot solve the symptoms that the endoscopist cannot understand. The flaw here is the direction of future AI development.
Nowadays, AI has emerged in the medical field. This technology will become a weapon to break the barriers of the traditional medical industry ; the break of this barrier will bring a huge innovation to the huge medical market.
Peeled Garlic,Peeled Garlic Cloves,Pre Peeled Garlic,Peeled Black Garlic
shandong changrong international trade co.,ltd. , https://www.changronggarliccn.com