Abstract:With the help of AI to improve the accuracy of medical imaging, we are like chasing the traces of an ant in the forest, and the tiny lung nodules have nowhere to hide under its diagnosis.
"With the help of AI to improve the accuracy of medical imaging, we are like chasing the traces of an ant in the forest, and the tiny lung nodules have nowhere to hide under our diagnosis. Yu Hong, director of the radiology department of Shanghai Chest Hospital, described it in an interview with reporters.
In fact, as early as 2016, Shanghai Chest Hospital launched an AI-assisted imaging diagnosis system to help radiologists capture the early microscopic changes of pulmonary nodules to help doctors make more accurate basic judgments, which is what we often call whether pulmonary nodules are benign or malignant. Through the continuous "feeding" of nearly 3 million real cases over the years, the AI system is also constantly learning and improving.
Zhu Lin, attending physician of the Department of Radiology of the Chest Hospital, told reporters that at present, under the accumulation of big data and AI-assisted analysis, doctors can do more innovative research in combination with clinical practice, such as evaluating the radiomics characteristics of patients' nodules, the difference in gene expression profiles and the heterogeneity of blood supply to patients' lesions, and the correlation between survival time, etc., which could not be studied and judged by the naked eye.
Suspicious nodules will not be let go
The accuracy rate of AI discovery has exceeded 99%
Chest hospitals were the first to use AI imaging technology to assist clinicians in making accurate diagnoses. Yu Hong, director of the radiology department of the Chest Hospital, revealed that with the continuous update and progress of the software over the years, the AI imaging technology has been constantly iterating, and in the words of red, at present, relying on AI reading, the accuracy rate of artificial intelligence to find pulmonary nodules has exceeded 99%, coupled with the doctor's second manual diagnosis, and strive to not let go of any suspicious nodules.
Yu Hong also showed reporters a lot of fine data of AI image analysis technology. In addition to tracking suspicious nodules layer by layer, AI also generates the size, density (calcification, solid, ground-glass, etc.) and morphological characteristics of the nodules, and can also preliminarily determine whether the suspicious nodules are high-risk or low-risk. Yu Hong said that when the patient is lying in the CT machine room for shooting, in fact, there is already AI technology to assist, some blurring caused by occlusion or metal tail shadow, etc., AI can already adjust it first, and make the image clear, and then provide it to the AI-assisted diagnosis system for follow-up analysis.
Yu Hong told reporters that the development of AI will also contribute to the development of more diagnostic technologies and indicators for early identification of lung cancer. Therefore, it is necessary to develop more accurate imaging detection methods for the diagnosis and prognosis of lung cancer. "At present, many scientific research projects in our department are trying to find early changes in lesions at the micro level through AI analysis, and these micro features can better play a role in suggesting lung cancer in the early stage. ”
The number of films seen by doctors can be multiplied
AI greatly improves work efficiency
During an interview with the radiology department, Jiang Yifeng, deputy chief physician of the radiology department of the Chest Hospital, told reporters that with the assistance of AI, the number of readings they can complete every day has increased significantly, and the work efficiency has been greatly improved. Many years ago, CT scans were often thick-slice scans, and doctors needed to look at a total of about 100-120 tomography images at a time to diagnose a patient's chest, and it would take at least about 10 minutes to complete the diagnosis of a patient. With the development of imaging technology, the chest CT examination in the chest hospital used thin-slice scanning as early as nearly 20 years ago, and the thickness of each layer of CT images is about 1 mm, and radiologists generally have to look at the mediastinal window and bone window images when making a diagnosis, so the number of pictures that radiologists need to review when diagnosing a patient is as high as 600-700, and the workload of doctors has increased significantly. With the assistance of the AI system, it can accurately circle the location of suspicious nodules in advance, and the doctor's reading efficiency is greatly improved, which also allows doctors to serve more patients every day.
Shen Yan, deputy chief physician of the radiology department of the Chest Hospital, said that the Shanghai Chest Hospital uses the AI system to diagnose up to 300,000 patients a year, which is among the highest in the world. Over the years, the AI system has been constantly being "taught" by doctors what may be malignant and what may be benign nodules, which has also made the progress of AI rapid.
"But with the current level of learning, it is still unrealistic to completely let it go to AI for diagnosis, for example, he will occasionally "make mistakes" and cannot identify nodules next to the mediastinum, etc., and we will continue to teach it in our daily clinical work. The accuracy of AI judgment is constantly improving, but for a patient, if we miss one and misjudge one, it will be 100% of the consequences for him, so our radiology doctors still need to be meticulous, and we still have to chase down the suspicious lesions. Chen Qunhui, deputy director of the radiology department of the chest hospital, said.
Under the feeding of big data
More innovative research is on the way
Zhu Li, deputy chief physician of the Department of Radiology, told reporters: "At the beginning, we needed to manually help AI learn, for example, at the beginning, the software could not even distinguish between blood vessels and nodules in the lungs. At first, it was up to people to teach the software, manually frame the nodule, mark it, and tell the software that it was a nodule. When the software learns a certain amount, it will automatically help us find the nodules. As shown by this software, it tells us where there is a nodule and what its density is, and it slowly learns how to analyze it, to do what a doctor does, and even to do a part of the work that surpasses that of a doctor. ”
The reporter learned that at present, in the chest hospital, many doctors are doing cutting-edge topics, Yu Lingming, deputy chief physician of the Department of Radiology, is trying to use AI technology to develop a more advanced auxiliary analog digital system, this innovative research uses AR reality augmented technology to build an AR glasses navigation system. Based on the patient's preoperative single CT scan to build an individualized digital twin model of the patient, the doctor can wear this special AR glasses to penetrate the patient's body and look directly at the digital organs in the patient's body, so he can directly guide the doctor to accurately reach the destination and release the marker with the puncture instrument in hand, and locate and mark the lung nodules or lesions of the lung area based on the patient's single preoperative CT scan.
In addition, Zhu Lin and other doctors are also using AI systems to conduct new research, "AI can support a lot of data mining research. For example, AI algorithms are used to study different clinical subtypes of lung adenocarcinoma, and the correlation between the different effectiveness of lung cancer treatment drugs, the survival of patients, and the microscopic imaging features of lung nodules is evaluated. Zhu Lin also mentioned that many new drugs have been developed clinically, and the microscopic morphological detection of lesions and organs through AI will be more conducive to evaluating some subtle changes in lesions and related organs after long-term drug use, and even predicting the appropriate drugs for lesions based on imaging features, and perhaps it is possible to find some drug side effects that have been ignored in the past in the future. "It's very interesting to do research in this area, but it's also quite difficult. ”
In the context of the all-round attack of AI, the advancement and innovation of science and technology are benefiting patients and protecting their health in a variety of rich forms.
![Ants lead elephants to dance, and kitchens find the strategic fulcrum of modern kitchens[fig]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)