Early detection and early diagnosis of pneumonia, AI gives doctors a pair of "eyes of fire"
Early detection, early diagnosis, and early treatment can significantly reduce the incidence and mortality of critically ill patients with new coronavirus infections. The combination of AI algorithms and doctors' experience will provide early detection and early diagnosis of new coronary pneumonia and even more types of pneumonia diseases. Efficient solution for early treatment.
"Science and technology of artificial intelligence (AI) diagnostic system in the detection of new coronavirus pneumonia and pneumonia differential diagnosis" and other 6 scientific research projects, recently received funding from the National Defense Science and Industry Bureau promotion special project and Suzhou emergency prevention technology special project channels.
Mining deep information that cannot be seen by the naked eye
According to the "New Coronary Virus Pneumonia Diagnosis and Treatment Program (Trial Version 7)" issued by the National Health and Health Commission, "Imaging Features" is listed as one of the three clinical manifestations of suspected cases of new coronary pneumonia. CT examination plays an important role in the diagnosis of new coronary pneumonia, and was once the main basis for clinical diagnosis in the main epidemic area. However, conventional CT examinations are also inadequate. It is difficult to observe relatively occult lesions at an early stage, and it is difficult to distinguish them from other viral pneumonia and bacterial pneumonia.
"Usually, image diagnostic physicians rely on the human eye to identify CT examination images, observe the images produced by the examination, and make subjective judgments based on their imaging performance and personal experience of the physician." Relevant project leader, General Nuclear Industry Hospital Fan Guohua, director of the imaging diagnosis department, said that this has certain limitations, there are many subjective influencing factors, and only some apparent image features can be interpreted.
Unlike doctors' naked eyes, artificial intelligence can transform visual image information into deep and characteristic information, and this information is quantifiable.
"This set of intelligent diagnostic systems established using artificial intelligence technology can detect the relatively early imaging changes that are not obvious to the naked eye, and can detect them. The second is that the qualitative changes are more accurate and the lesions can be more accurate. Diagnosis; in addition, the time consuming of the entire process can be greatly reduced compared to manual labor. "Fan Guohua said, for example, an adult who performs a chest CT examination will produce four or five hundred thin-layer images. It is time-consuming and labor-intensive to look at each one manually. But the machine can detect these four or five hundred images in a few seconds, and there is no problem of fatigue caused by continuous work.
Obtaining standardized image data for large samples is difficult
Use artificial intelligence technology to perform deeper analysis on image data. The specific process is to obtain digital images through CT scans, and then import the image data into the software system for analysis, and build models through machine "deep learning". For the established model, a certain number of confirmed cases are used to verify its reliability, and then it is used to detect other unknown cases. This combination of artificial intelligence and imaging diagnosis has previously been used for the diagnosis of tumors.
To "grow an adult", eventually mature to be able to help diagnose, the "nutrition" to feed it for learning and training is data.
"The larger the sample size, the more standardized data the better. But as the research progresses, to increase the sample size, the data acquisition of a single hospital is limited, and coordinated multi-center research is needed to expand the sample size." Fan Guohua said Sample standardized image data is difficult.
When performing a CT examination, the equipment and scanning parameters used by each hospital are different. However, for large sample analysis, all image data are required to be standardized and standard. Before the image data is handed over to the machine, the data needs to be labeled. Because of this, the accurate labeling of the data has an important impact on artificial intelligence applications. This means that to accurately segment the lesion, "usually the more accurate the segmentation, the better, but it is also a difficult point." Fan Guohua said.
After the outbreak, it will be used for the differential diagnosis of pneumonia
In recent years, the combination of computer technology and imaging diagnosis has become increasingly close. Fan Guohua said that the starting point for carrying out this research is to provide some help for clinical diagnosis and explore earlier and more accurate diagnostic methods. "Currently, the optimization of CT scanning technology has included work on standardization of data acquisition, while collecting imaging data, clinical data, laboratory test data, etc. of related cases, and then using these data for modeling." Fan Guohua said that the research is expected It will be completed in about one year, and it can be put into clinical use from January to February next year.
If the new crown pneumonia epidemic has passed, will this result be useful?
"We hope to get practical application as soon as possible." Fan Guohua said that different pathogenic microorganisms can cause inflammation of the lungs. After the epidemic, this diagnostic system will be mainly used for the detection and differential diagnosis of pneumonia, and for some people who need attention. Lung inflammation.
At the same time, he said cautiously that the system is still in the research process, and it is still in the early stage of research. What level can be reached in the future depends on the development of future work.
Despite the advantages of artificial intelligence, Fan Guohua believes that machines cannot completely replace human roles. (Li Chunping reporter Chen Yu)