Just a few days ago, in the Cancer Radiology Department of a hospital in Beijing, Dr. Wu imported a lung scan of a patient suspected of lung cancer into a software that looks like Photoshop. This is actually an artificial intelligence algorithm-based software that can use a neural network trained by thousands of scans to mark the suspicious nodules in the scan results with a red box.
Dr. Wu carefully examined the results of the software processing. She corrected two false positives. The neural network misidentified the blood vessels as a tumor. But the neural network also marked a nodule she had missed before. This nodule may be a sign of early disease.
At present, China is vigorously advancing the application of artificial intelligence algorithms in the medical field. The United States and Europe also have such trends, but China has fewer restrictions on data and new technologies, and the need for automation is even more urgent. There are about 1.5 doctors per 1,000 people in China, and 2.5 in the United States.
China’s progress in this area is very rapid. According to Yiou Intelligence, a consulting company in Beijing, about 130 companies in China are developing artificial intelligence algorithms and applying them to the medical field.
From next month, a hospital in Beijing will also use artificial intelligence algorithms to process all lung scan data, thereby accelerating the process of disease screening.
Last year, the Ministry of Industry and Information Technology issued an artificial intelligence three-year plan to strive for an important breakthrough in 2020, and the application of artificial intelligence in the medical field is also an important part of the plan. Last month, a report released by IDC predicted that by 2020 China’s artificial intelligence medical service market will reach 5.9 billion yuan. Not only is it a startup company, China’s giants of technology companies have also joined the battle. Alibaba and Tencent have R&D centers to develop artificial intelligence diagnostic tools.
In addition to policy orientation, the Chinese public’s attitudes and attitudes toward artificial intelligence may also make China less vulnerable to the landing of artificial intelligence than Western countries, and it is easier to use artificial intelligence to advance the healthy development of the medical industry. The rise of artificial intelligence in Western countries has always been a concern for doctors, but most Chinese doctors are very keen to use software to automate most of the repetitive tasks.
Despite this, artificial intelligence still has significant obstacles in the medical field. The conclusions of such neural networks through complex computational processes are often difficult for humans to interpret. This leads to many problems, such as the use of artificial intelligence for medical diagnosis. Who should be responsible for the accidents that occurred during the process.
The development of algorithms that can process CT, X-ray, and other medical image data is a hot area in China. The pursuit of imaging in the field of imaging is due to the fact that the technology of using artificial intelligence for image recognition is relatively mature at this stage. Dr. Wu’s artificial intelligence software was developed by a startup company in Beijing. More than 20 hospitals in China have installed this software. The company has collected data from more than 180 hospitals and formed a large network.
However, artificial intelligence can also be applied to other areas of the medical industry. Dr. Lu is an oral rehabilitation doctor in Beijing. He is working with Tsinghua University to develop an artificial intelligence algorithm that can design dentures. They used the principles of denture design derived from textbooks and 30,000 real cases that doctors had labeled to train algorithms. Dr. Lu said that the dentures designed by this algorithm are comparable to experienced doctors. So he also plans to apply for evaluation of the algorithm’s clinical trials in the second half of this year.
Dr. Liu from Beijing is a doctor who specializes in lymphoma. He is working with researchers at Tsinghua University to develop machine learning algorithms to analyze ultrasound data to detect treatment-induced thrombosis in lymphoma patients. If you can use ultrasound scanning to identify patients who are still at an early stage of venous thrombosis, the patient can be treated promptly. However, there is also a problem that is more difficult to solve. The hospital does not have enough resources to screen every patient. Usually, the patient is only tested when symptoms appear.
Some researchers in China are also studying how to make these software automatically learn medical knowledge. Xiaomei, which was jointly developed by HKUST and Tsinghua University last year, successfully passed the National Physician License Examination and scored more than 96% of the candidates. In fact, the difficulty of creating this kind of algorithm is not in the knowledge of expanding the software, but it is necessary to teach the machine how to understand the internal relations of different knowledge points, and apply it to the reasoning and decision-making process.
The core of this type of algorithm is still the processing of natural language, but more focused on the medical field. This kind of algorithm is totally different from the process of human choosing the correct answer when answering multiple-choice questions. The algorithm finds the evidence needed to answer a particular question by calculating the statistical similarities between the words.
After a detailed analysis of the examination results, we found some places where machines could not compete with humans: Common sense and ethical issues. Machine algorithms answer the questions in these areas with scores lower than the national average.
Prof. Wu from Tsinghua University dominated the research project and he is now exploring ways to apply this algorithm to the clinic. He also admitted that this approach would not be as simple as installing the software on every doctor’s computer.
But on the other hand, doctors who have actually used these softwares think that these tools are helpful for their work. For example, Dr. Wu’s hospital has about 10,000 outpatient visits per day, so she actually does not have enough time to see each film carefully, so these software can really help them relieve some of the burden, she said.