On June 10th, the “Artificial Intelligence and Pathological Section Diagnosis Symposium” hosted by CCF YOCSEF was held in Shanghai. Doctors, AI experts, investors, lawyers and other heroes gathered here to discuss the problems in building intelligent medical products.
Giiso Information, founded in 2013, is a leading technology provider in the field of “artificial intelligence + information” in China, with top technologies in big data mining, intelligent semantics, knowledge mapping and other fields. At the same time, its research and development products include information robot, editing robot, writing robot and other artificial intelligence products! With its strong technical strength, the company has received angel round investment at the beginning of its establishment, and received pre-A round investment of $5 million from GSR Venture Capital in August 2015.
I. At present, is artificial intelligence diagnosis reliable?
“I don’t know if the AI diagnosis is reliable, but based on the Google results, I don’t think it is.” Professor Zhu Hongguang joked that the diagnostic accuracy rate of 88.5% is to die, was killed by others, Professor Zhu Hongguang is the International Society of Pathology In China, vice chairman of the Department of pathology of Fudan University basic Medical School.
In March 2017, an AI developed by Google, Google Brain and Verily scientists to diagnose breast cancer competed with pathologists. The pathologists scored 73.3 percent of the time, while the AI scored 88.5 percent, which led to media reports that AI “beat humans”.
Professor Zhu said that if the AI system misdiagnoses grade 2 ductal carcinoma as grade 1, the treatment methods are similar, but the diagnosis of cancer cannot be wrong, and any wrong diagnosis is a medical error. “With ai diagnosis, it’s not plausible, at least today.”
What is the bottleneck of ARTIFICIAL intelligence diagnosis?
“It’s a difficult question to answer, but I don’t think there’s a good look at the totality of human intelligent medicine right now.” Professor Zhu said, for example, a slice, the first time, judged as malignant, to high magnification of the local turned benign, but the final conclusion is malignant, because the doctor’s judgment is not based on cells, but biological cell behavior. Pathologists focus on low magnification, because low magnification can see the whole picture.
In response, Dr. Li Guannan, founder and CEO of Weview Intelligence, said that the algorithm has corresponding processing results at different scales. They not only look at local areas, but also look at the whole. “THE AI is only responsible for finding the lesions, and the final determination is done by doctors.” He shared that the company’s current programs focus more on scientific research than on clinical diagnosis. In the early stage, most of the data were processed around cells, and in the later stage, it was slowly transferred to the analysis of the whole region, including the texture changes of tissue areas and the impact on the work of pathologists.
HE Jin, a member of the Standing Committee of the Special Committee for Pathological Technology and Equipment, believes that there are three bottlenecks restricting artificial intelligence diagnosis: First, standardization of sections: HE sections are the basis of pathological diagnosis, and the thickness, quality and staining quality of sections should be guaranteed. If the standards are not unified, the final results will be bad. Second, the definition of the image; Third, AI companies should seek closer cooperation with pathologists.
Iii. How to combine ARTIFICIAL intelligence and medical care?
Dr. Tao Xiaodong, General manager of IFLYTEK Intelligent Medical Business Division, believes that the combination of ARTIFICIAL intelligence and imaging can solve these problems: first, things that doctors do not have time to do, or things that are a waste of time for experts, but are very necessary for diagnosis, such as making records when taking samples, image screening, etc. Second, the things that computers are better at doing, doctors are better at qualitative diagnosis, but computers are better at quantitative diagnosis; Third, it is more important for radiology department and radiology department to solve the problems of insufficient technical experience and inconsistent imaging standards. For pathologists, computers help make imaging more standard.
What does he think AI could be most useful for medicine? This is a technical problem, but also an application problem, AI experts need to communicate with medical professionals, spark collision.
Yu Guanzhen, deputy chief physician of the Oncology Department of Longhua Hospital affiliated to Shanghai University of Traditional Chinese Medicine and a distinguished professor, said she is concerned about whether diseases that are difficult for pathologists to diagnose can be achieved through new means and technologies. Li guannan believes that the main role of artificial intelligence is to help pathologists reduce their workload and locate lesions first.
Giiso information, founded in 2013, is the first domestic high-tech enterprise focusing on the research and development of intelligent information processing technology and the development and operation of core software for writing robots. At the beginning of its establishment, the company received angel round investment, and in August 2015, GSR Venture Capital received $5 million pre-A round of investment.
4. Who owns the data?
Li Gang, a senior partner at Beijing Yingke (Shanghai) Law firm, said that currently, the management and implementation of medical records in medical institutions is based on the health and Family Planning Commission’s regulations and is confined to medical institutions. At present, there are no clear legal provisions on the ownership of data, but patients are the producers of cases, and hospitals have intellectual property rights. Is the medical record shared by patients and hospitals? There is no clear definition yet.
According to Lei Feng, article 42 of the Cyber Security Law, which took effect in June, says that if the data has been processed and irreversible, it can be used, but infringing personal privacy through some technical means is problematic. The so-called personal information, there are two aspects, one is identity information, such as occupation, name and so on; Second, life trajectory information.