"Chinese lung nodules are not enough."
When I learned that PACS (Picture Archiving and Communication Systems) of a famous top three hospital in Southwest China had access to the intelligent imaging diagnosis system for lung nodules in seven artificial intelligence companies, an imaging doctor sneered at the current situation.
Lu Xun once said that there is no road in the world, and there are more people to go. Naturally, there is a way. But in the detection of artificial intelligence in lung nodules, Chinese start-ups have already stepped on a Beijing-Shanghai high-speed. Many AI manufacturers are fiercely competitive in core hospitals. It is common for a radiology department to put 4-5 AI vendors' servers.
As of July 2018, incomplete statistics, in the field of lung nodule screening, there are more than 20 artificial intelligence companies that have produced specific products, most of which have obtained venture capital, and more returnees and research institutes. The team of hospital experts is gearing up and eager to try.
Everyone is heading for the same goal.
Who brought the fire to the lung nodules?
From the demand side, the number of new lung cancer patients in China is the highest in the world, the number of lung cancer deaths is the highest in the world, the demand for early screening is strong, and low-dose spiral CT is being widely promoted. From the perspective of image quality, chest CT images Layered thin, clear vision, less interference factors, and rule of disease characteristics can be used as an ideal place for intelligent image interpretation. In addition to the scarcity of Chinese imaging physicians and the promotion of national policies, the application base in this field is perfect.
However, the demand for any disease sector in China will be extremely large due to the huge population base. It is the era of big data that really makes the lung nodule intelligence detect fire.
In July 2017, Pranav Rajpurkar, head of SQuAD, unveiled his algorithm for chest x-ray pneumonia and published the world's largest publicly available chest x-ray dataset, which contains 10 of 14 diseases. More than 10,000 x-ray main views.
In October 2017, the NIH (National Institutes of Health) Clinical Center released an x-ray image containing more than 30,000 patients, more than 112,000 frontal lines of sight, and 14 categories excavated from radiology reports using NLP technology. Image label for disease, free for use by researchers worldwide.
In July 2018, the NIH (National Institutes of Health) Clinical Center once again shared a large CT image database containing 32,000 CT images and disease imaging data to help scientists and clinicians enhance their disease imaging. To learn the diagnostic skills, AI developers can also freely use the above data for the training of the AI ​​system.
For entrepreneurs in China's medical artificial intelligence companies, these US public data sets are tantamount to the most comfortable cradle, providing entrepreneurs with a wealth of data "food"; the accuracy of the AI ​​model is trained through public data sets. After reaching a certain level, it will be put into clinical practice, and the “feeding†model of massive CT images in China will make its performance leap further and become the universal path of Chinese medical AI enterprises.
“For entrepreneurs, the most difficult thing to get is clinical medical data. No matter how strong the technical foundation is, there are many algorithm models available on the market, there is no clinical data to train AI, everything is passive water. The larger CT data base in China makes the progress of China's artificial intelligence technology no less than its birthplace - the United States." An image cloud operator said so. In an article in the 2018, the New York Times (NYT) mentioned that China's Ali and Etu have already introduced AI to the medical industry before Amazon.
But at the same time, he also reminded that although the starting track of domestic AI entrepreneurs is similar, it does not mean long-term competition in the same dimension.
"As the achievement of 10 years after graduation, there is a world of difference. The same is the choice of the intelligence detection of lung nodules as a breakthrough, the strength of research and development, the depth of understanding of the medical industry, the foresight of strategic vision, the size of the industrial structure, Will affect the development prospects of AI companies." The manager believes.
The evolutionary path of pulmonary nodules AI
In 2017, the major AI companies detected in the main lung nodules have delivered brilliant answers, and the sensitivity has soared, 95%, 96.5%, 98.8%... pixel differences that are difficult for human eyes to detect. AI's powerful computing power is invisible.
However, after all, medicine does not only understand the basic function to solve the problem.
At present, in order to ensure the accuracy of image interpretation, a licensed doctor and a deputy chief physician usually read the chest radiograph of the same patient. After the radiographer finishes reading, the doctor of the same level needs to review it again and sign it. The purpose of AI is to replace the first step in this step, because AI doctor not only has excellent "vision", but also can see almost every tiny nodule. At the same time, AI doctor is tireless, does not have visual fatigue, and sees thousands. Tens of thousands of chest CTs are also between milliseconds.
The sensitivity of AI is constantly increasing. In theory, AI can find every nodule in the lungs, but the false positive rate that comes with it is a headache. It is to quickly raise the sensitivity to the extreme and temporarily ignore the false positive rate. Still spending more effort to increase sensitivity and false positives to usable levels? Or is there a more scientific and accurate assessment indicator?
Professor Gong Xiangyang, a well-known domestic radiologist and director of the Department of Radiology of the People's Hospital of Zhejiang Province, once said that it is difficult to take into account both specificity and sensitivity. Therefore, many companies will give priority to sensitivity when developing systems. Under the premise, improve specificity.
“Sensitivity and false positives are really difficult to achieve. They are extremely challenging to test the technical skills of AI companies. They must maintain sufficient sensitivity, but also ensure a sufficiently low false positive rate to ensure that most of the identified nodules are Correct, it is clinically meaningful, and there is no such thing as the outside world seems to be so difficult. According to the hard work of the medical care, it has achieved the industry leading level in both aspects," Ni Hao, the president of Yitu Medical, was interviewed by the media. Said, "If the false positives are too high, it will greatly increase the doctor's work pressure, and also lose the original intention of AI to help the clinic."
Ni Hao also reminded that simply paying attention to sensitivity and neglecting the false positive rate will not only increase the workload of doctors to check image reports, but also cause great psychological pressure on patients, and even lead to over-diagnosis under panic, waste medical resources and increase patients. burden.
In order to more objectively reflect the promotion of AI products for clinical work, Yitu Medical introduced a new measurement standard in the industry - the structural adoption rate of structured reports.
Specifically, the indicator consists of two aspects – “structured report†and “clinical adoption rateâ€. The “Structural Report†not only requires the lung nodules AI to find nodules, but also the information on nodule size, trait description, benign and malignant signs, and a structured clinical report; and the “clinical adoption rate†is more demanding. - How much of the structured report generated by AI can be directly adopted by clinicians without modification?
Yitu Medical announced their care. The clinical feedback of the aiTM lung cancer imaging intelligent diagnosis system in the first half of 2018 showed a clinical manifestation of 92%. This is an extremely excellent performance, meaning that 92% of the structured reports produced by the AI ​​system are directly accepted by the imaging physician. Behind this number, the working time saved by the AI ​​system for the imaging physician is difficult to estimate.
"Only artificial intelligence embedded clinical workflow of doctors, especially doctors recognize our test report, AI to enhance the efficiency of clinical work is meaningful, sensitive indicators is very important, but the sensitivity is only the beginning," according to the medical chart Zheng Yongsheng, vice president of products , said, "At present, with this unique technology, the system has not only entered more than 100 top three hospitals nationwide, but is also fully rolled out in the grassroots hospitals where AI Medical really needs to be empowered."
Breaking through the ceiling of the lung nodules
When the detection rate of lung nodules gradually approaches the theoretical limit, all medical AI companies are thinking, where is the next step in the intelligent diagnosis of chest CT?
The essence of this question is asking how the AI ​​ability is advanced, from answering "what to see" to answering "what" and "how to rule."
"On the one hand, we continue to dig deep into the nodules of the lungs, from image interpretation to MDT decision-making. On the other hand, from a single pulmonary nodule, we detect the intelligence of various diseases such as pneumonia, tuberculosis, chronic obstructive pulmonary disease and bronchiectasis. Diagnosis, solving the problem of departmental scene with a single application, jumping out of a disease, an AI," said Ni Hao. "This road will be very long and difficult, but only if you continue to meet the clinical scene, you can truly become a good helper for doctors. To promote the construction of smart hospitals in the future."
On June 15th, Yitu Medical and China's top three hospitals, Huaxi Hospital, released the world's first multidisciplinary intelligent diagnosis system for lung cancer. This system is called “the most doctor-thinking†AI application, which not only enables nodule screening. Such as the primary function, it can achieve the diagnosis coverage of all types of lung cancer lesions, comprehensive multidisciplinary clinical information for comprehensive diagnosis, the decision-making basis is based on the latest international and domestic clinical lung cancer diagnosis and treatment guidelines, and with the increase in the number of clinical treatments, the more The smarter and smarter you are, the more effective it is for grassroots physicians to improve the level of lung cancer diagnosis and treatment and reduce misdiagnosis and missed diagnosis.
What is more gratifying is that some of the functions of this product have already begun clinical trials and have undergone severe clinical tests. If successful, it will be a great improvement in medical AI.
"Like all the technological changes that have been experienced in human history in the past, artificial intelligence will also be integrated into the doctor's work process, and together with the doctors' group to better benefit the majority of patients", China Medical Imaging AI Industry and Education Research Innovation Alliance Chairman, Professor Liu Shiyuan, director of the Department of Imaging Medicine and Nuclear Medicine at Changzheng Hospital of the Second Military Medical University, said that “the future is here, but the distribution is uneven.â€
We're professional Monopolar Forceps manufacturers and suppliers in China, specialized in providing high quality medical instruments with reasonable price. We warmly welcome you to buy or wholesale bulk forceps for sale here and get quotation from our factory.
Monopolar Forceps,Laparoscopic Forceps,Laparoscopic Grasping Forceps,Laparoscopic Dissecting Forceps
Tonglu WANHE Medical Instrument Co., Ltd , https://www.tlvanhurhealth.com