What kind of AI project attracts hospital payment?
Can a new generation of artificial intelligence set off an industrial revolution? This is not only a problem for AI development companies, but also a concern for beneficiaries such as doctors and patients.
In the afternoon of April 12th, Tongda Capital hosted the 12th salon of Tongdu Time co-organized by Arterial Network and Tsinghua Institute of Economics and Management. “Where is the 'Medical + AI' road?†Application value and commercialization of a new generation of artificial intelligence in the medical field The potential has entered into in-depth discussions.
The meeting invited Professor Song Sen of the Department of Biomedical Engineering and Brain-like Research Center of Tsinghua University; Professor of the First Maternal and Child Health Hospital in Shanghai; Duan Tao, founder of Chuntian Medical Management; and Li Yiming, founder of Shen Rui Medical; Li Xing; Li Xiaodong, co-founder of Lianxin Medical; Zhang Dadi, partner of Danhua Capital; Fei Xiaoyan, chief engineer of the Xuanwu Hospital Information Center, and Yu Hui, partner of China Electric Power Fund, were the guests of the salon. At the same time, numerous industry experts, investors and entrepreneurs participated in this event.
"Medical + AI" should go from here, let us find the answer from the content of the meeting.
Using images to find lesions
Image inspection is the most common method for patients to go to hospital for examination. At this stage, medical imaging equipment is very popular. At present, 80% of clinical treatments require prior CT and MRI examinations. The data produced by these devices is standardized and very easy to handle.
However, in daily work, radiologists generally spend a lot of time browsing medical images. The browsing process of each image is generally only about five minutes. Repeated browsing and identification for a long time will reduce the doctor's correct rate. Therefore, misdiagnosis often occurs.
Deepin Medical Co-founder Li Yiming pointed out at the meeting: "In today's healthcare system, we do not have enough resources to support a large number of high-quality medical resource needs. This is the opportunity of 'Medical + AI' to make up for the current situation through AI. With a huge gap, Shenrui Medical will use medical imaging as an entry point to gradually explore the value of 'medical + AI'."
Many doctors are reluctant to put their energy into the examination of healthy people. They can't provide timely services to the sick people and they don't have time to serve healthy people. Take lung CT as an example. If the CT is requested at the outpatient clinic, the doctor will often be very careful. However, if the CT is done during the physical examination, the doctor may not take the time to browse in detail, so down, small lesions are easily overlooked.
Pulmonary nodules detected (picture from Deepin Medical)
AI technology can play a great role in the image field. On the one hand, the image data structure is standard and the data volume is large, which makes it easy for AI learning. On the other hand, AI does not feel sleepy because of repeated labor, even if AI has misunderstood the image. However, the actual miscarriage rate is much lower than that of a radiologist.
AI can free doctors from repeated low levels of labor liberation and allow doctors to do more meaningful things. In the future, patients need more humanistic care. This is what cold machines can't do.
Can AI Solve Pain in Medical Information
There are three problems with medical information: incomplete, inaccurate, and unstructured.
Duan Tao, the founder of Chunyu Medical Management, discussed the medical information issue at this meeting. He believes that the root cause of the above problems in medical information comes from insufficient medical resources, and too few high-quality medical resources have taken too much burden. Although the Wei-Qiang Committee distributed enough forms to the doctors to collect information, the doctors are always in a state of high workload and cannot even cope with work. How do you spend time sorting out medical data? This led to the doctor's perfunctory of the corresponding form, the form lost its own meaning, and the data collected was also incomplete and useless data.
AI can solve this problem from both indirect and direct approaches.
On the one hand, AI can directly analyze the condition and judge the patient's physical condition. In this case, AI requires two kinds of data—patient genotypic data and phenotypic data. The hospital can hand over the patient's case to the AI ​​in advance. After receiving the genotype data, the AI ​​can receive the patient's own input or the phenotypic data obtained by the doctor in person to determine the condition.
In this case, the doctor can save a lot of time, and the confirmed data is also standardized, structured, complete and accurate. In this way, AI can release doctors' time and resolve pain points in medical data.
There is bound to be a false positive rate for AI-assisted analysis, but we must have enough patience with the machine because even a senior doctor can misjudge for various reasons. According to Dr. Duan Tao, for an experiment in 2017, an AI system defeated 95% of doctors in analyzing the condition.
On the other hand, AI can label and analyze images and other data; automate the processing of information generated by medical devices; organize and analyze large data according to algorithms; these functions will reduce the burden on doctors, and doctors have more energy to analyze. Some data that cannot be processed by AI indirectly solve the pain points above medical information.
How does the AI ​​field develop in the future?
Prof. Song Sen from the Department of Biomedical Engineering at Tsinghua University told the conference that he had learned from Tsinghua's graduation to MIT. And it leads to two major problems, all related to the future development of AI.
1. How to use AI technology to learn small samples?
Professor Song Sen believes that so far there are many diseases that belong to minor diseases, but these diseases have a high fatality rate. Now medical technology is difficult to overcome, and the data collected is not as good as common diseases.
In spite of so many problems, Professor Song Sen believes that these problems are not without solution. In fact, we can use the interactive markup method to let AI learn with big data, train with small data, and gradually find the law from it.
2. How does AI make decisions?
In the course of learning and analysis, human beings always quantify various indicators. Even decisions based on experience must be derived from the revelation of something far from the past, rather than being produced out of thin air. AI can make decisions based on big data for learning and analysis. However, throughout the entire process, researchers have no way of knowing how AI makes decisions. The AI ​​analysis process is like the researchers putting data on one side of a black box and then taking the results from the other side. As for what happened in the middle, no one knows.
However, if people can understand how AI makes decisions, it is possible to obtain new conclusions by analyzing these thinking methods and to improve the deficiencies in existing AI calculations. Professor Song Sen believes that if we can see the problem from the perspective of AI and understand how AI thinks, we can make rapid progress in medical technology. Let AI have both intuition and reasoning about the process of medical analysis.
Based on deep learning and probability-based Bayesian model to provide basis for AI decision (from the same capital)
Investor: Profit is the motivation for investors to participate
How AI's business approach lands is a matter of most concern to investors. AI's talent is very expensive, and AI companies have to constantly develop products, expand appointment channels, and continue to recruit talent. However, there is no clear way of income on the income side. Many AI companies can only rely on continuous financing to survive in the first few years, and their losses will increase.
Even so, Yu Hui, a partner of CLP Health Fund, believes: "One is that the medical industry will not develop so fast. It is very difficult to see a good business profit model in the short term. But in the long term, the third quarter of 2018 or four In the quarter, there will be favorable news in this industry. The trend in 2018 will be upward, and it may reach 2019, and even 2020 will continue to maintain upward trend."
Zhang Dadi, partner of Danhua Capital, thinks: “Exploring the business model of the AI ​​project is not only a problem in China but also a problem in the world. China has such good source data, and the country certainly does not want to lag behind in artificial intelligence. In the planning process, the country has written AI into its national development strategy, and few countries do so. Therefore, our country will not only follow the pace of the FDA, but also accelerate the development of AI. This is what we are currently doing. For AI's start-up companies, it is a very good news. As long as it is done, the value will be reflected."
Zhang Dadi said in an interview with the Art Network reporter: “We don’t look at which area the technology belongs to, AI is good, or the blockchain is good. We will only invest in projects that may be profitable in the future.†So even if the AI ​​looks like the future But if investors can't find a good direction to solve the profit problem, investors will gradually lose their confidence.
Hospital: What kind of project is worth our payment?
Xu Xiaoyan, chief engineer of the Xuanwu Hospital Information Center, pointed out some practical problems facing the current AI at the round table discussion stage of the conference. She believes that the biggest reason why AI still cannot land is still in technology. Taking the hospital image as an example, an AI can only screen certain diseases, but in practice, the doctor gets a CT and will analyze it comprehensively instead of thinking in terms of preconceptions. — First learn that the patient has a certain disease, and then use the corresponding AI for detection. It is unreasonable for the hospital not to have enough funds to deploy enough AI. It is also an issue that cannot be solved now.
Therefore, hospitals in the subdivided areas have high demand for AI, but for large comprehensive hospitals, AI at this stage cannot meet the needs of the hospital. “As long as AI can achieve the level of neurology in our hospital, we will not take the initiative to raise it, and we will also take the initiative to promote it. This is possible. But I think that this also requires technology companies to continue their efforts to find the core of the problem. â€
The development of new things is always twists and turns, but there is no doubt that AI, as an extension of human wisdom, has gained people's recognition, it is only a matter of time. We need to have confidence in the AI ​​industry, be interested in AI developers, and one day we can see our dream science and technology world.
Guangdong Kaihua Electric Appliance Co., Ltd. , https://www.kaihuacable.com