The diagnosis result of refined AI of ＂master film reader＂ is close to the level of senior professional doctors.

　　Surprised or not? Your CT film is watched by an artificial intelligence computer!

　　The Medical Imaging Department of Nanjing Gulou Hospital applies the fashionable AI technology to the field of medical imaging. For patients undergoing chest CT examination in Gulou Hospital, the first "reader" of CT films is a specially trained artificial intelligence robot, which greatly improves the accuracy and speed of reading films. Recently, EBiomedicine, a sub-issue of The Lancet, introduced in detail the clinical application and transformation results of the SCI paper on "Artificial Intelligence+Medicine" led by Zhang Bing, director of the Medical Imaging Department of Nanjing Gulou Hospital. (Correspondent Liu Huiyan Intern Lu Wenqian Yangzi Evening News/Yangyan reporter Cai Yunqi)

　　A magical "game"

　　It only takes a second for a doctor to look at hundreds of pictures with the naked eye.

　　Director Zhang Bing has three computers on her desk. She demonstrated a naked-eye PK artificial intelligence reading competition to reporters. Director Zhang Bing first observed with the traditional naked eye mode to find lung nodules on CT films. There are more than 300 CT images of this patient. After preliminary screening, it is found that there is no big space occupation. However, if you want to observe small nodules, you still need to look at these more than 300 images layer by layer. Director Zhang read more than 300 pictures layer by layer on the computer, and it took about 10 seconds to preliminarily detect a small nodule with the naked eye. Then, on the other side of the computer, Director Zhang Bing calmly turned on the artificial intelligence system, just clicking the mouse on the image of the artificial intelligence system, and the diagnosis results popped up instantly on the work page: 9 suspected lesions. On the image on the left side of the work page, each suspected lesion is circled with a colored box and marked with a serial number. At the corresponding position on the right side of the page, each suspected lesion is listed according to the serial number, and the benign and malignant nodules are judged. The reporter saw that the first nodule detected by artificial intelligence coincided with the nodule observed by Director Zhang Bing with the naked eye. From the second, it was a nodule that was very small, light in color and difficult to find with the naked eye.

　　Speaking of the magic of artificial intelligence, Director Zhang remembered a case that impressed her deeply. A patient’s chest CT was examined by a doctor and found multiple small nodules, which could not be counted, but artificial intelligence accurately reported more than 800 nodules and clearly located the malignant nodules. This shocked everyone!

　　The "master reader" is tempered like this.

　　First-class experts as teachers, "feed" the data of hundreds of thousands of patients in TA.

　　How did the artificial intelligence computer master this stunt? Director Zhang Bing introduced that artificial intelligence has gone a long way before it is really put into clinical application. It takes more than three years to prepare the whole system and more than two years to turn the results into products. The whole team’s training of artificial intelligence is like a group of teachers educating and training a child to take the exam. In order to keep the children from making mistakes in the exam, the whole team fed a lot of "exercises" to the "children" — — Data. For example, in the practice of diagnosing pulmonary nodules, first-class experts in the field should be gathered. They input a lot of data about the diagnosis of pulmonary nodules to artificial intelligence, so that TA can learn what nodules are in memory again and again. In order to make the learning effect better and more rigorous, it is necessary to form a professional correction team to make rigorous feedback on every judgment of artificial intelligence. Only when more than two experts give the same opinion, will they give this answer back to artificial intelligence, so that artificial intelligence can learn deeply in constant practice and constant feedback, so as to master this skill thoroughly.

　　Can artificial intelligence make mistakes? Director Zhang Bing said that you don’t have to worry about it. Relevant data show that the diagnostic accuracy of artificial intelligence is very high, which is higher than that of junior doctors and close to that of senior doctors. Moreover, artificial intelligence is only the first level of film reading, as well as the second reading and review by doctors with senior titles, and the final report is the responsibility of the doctors who review the report.

　　The doctor’s two-hour workload was shortened to 100 seconds.

　　Director Zhang Bing introduced that the application of artificial intelligence in medical imaging brings all-round benefits to doctors, patients and the treatment process.

　　In the traditional artificial mode, imaging department is a platform to support the discipline, and doctors need to face a lot of workload every day. Take the imaging department of Gulou Hospital as an example, it needs to receive 2500 patients every day on average, while a normal person needs to see at least 300 images in CT. Taking the diagnosis of pulmonary nodules as an example, the size, shape and blood vessel section of nodules are very similar. Doctors in the traditional mode need to look at tens of thousands of images every day, and long-term work will inevitably lead to eye fatigue and mental inability to concentrate. Therefore, it is inevitable for doctors in imaging department to miss the diagnosis of nodules, which is also one of the pain points in the workflow of imaging department. After the introduction of artificial intelligence, tens of thousands of images have been handed over to artificial intelligence, which used to take 2 hours, but artificial intelligence can detect very accurate results with a click, which greatly improves the diagnostic efficiency.

　　There are about 300 chest CT images of a normal adult, but it is impossible to print all these 300 images in the actual printing process. Generally, one is printed every four images, so about 80% of the images are lost in the printing, and there are likely to be very critical images in the lost images, which is another pain point in the workflow. For example, a patient’s report shows that there is a big nodule on the 160th layer image, but the image on this layer is just skipped when choosing to print the image. To solve this problem, it is necessary to find the 160th floor again and print it again, which is very inconvenient for doctors and patients. Artificial intelligence automatically identifies and collects key images with nodules, and automatically puts these key images on the upper layer when printing and typesetting, which overcomes this difficulty well.

　　Extended reading

　　Will AI replace doctors? The answer is no.

　　Talking about whether artificial intelligence (AI) will completely replace the role of doctors in imaging department in the future, Director Zhang Bing thinks it will not. She laughed and said that artificial intelligence AI is still only the level of primary school students, and can only replace some simple repetitive work.

　　Although artificial intelligence has greatly reduced the workload of doctors and helped them to complete most of the work, these jobs are relatively primary and simple, really complicated and need comprehensive thinking to judge, and artificial intelligence can’t replace them.

　　In the future development, artificial intelligence will mainly assist doctors to solve many simple and repetitive tasks, so that doctors can focus more on the diagnosis and treatment of intractable diseases.

　　More applications in the future: it is expected to screen out tumors at a glance.

　　Artificial intelligence technology is the trend of scientific and technological development in recent years, and the combination of artificial intelligence and medicine is also the direction of medical development. Nowadays, the application of artificial intelligence is only limited to the judgment of a single disease, and it is more common in the diagnosis of pulmonary nodules. Director Zhang Bing told the reporter that her team has been deeply cooperating with Guo Hongqian, director of urology department of Gulou Hospital, to seek enough data to find prostate cancer in the process of image observation, so as to avoid the in-vivo puncture test that made patients very painful in the past.

　　"I hope that the future artificial intelligence will continue to grow under the support of sufficient data and sufficient training feedback, and it will be able to cope with more complex diseases, and play a role in improving efficiency and reducing errors in the whole process of imaging, so as to better help doctors treat patients and benefit mankind." Director Zhang Bing said. (Liu Huiyan, Lu Wenqian, Cai Yunqi)