Chair Professor Zhang Kang
Since the outbreak of the new crown pneumonia at the end of 2019, the Faculty of Medicine and HKUST Hospital of Macau University of Science and Technology have been actively engaged in research on the epidemic. In April 2020, based on 500,000 chest CT images, it successfully developed an artificial intelligence imaging_assisted diagnosis system for new coronary pneumonia, with an accuracy of more than 90%. It has been deployed and used by multiple units internationally and has been highly praised and affirmed by all walks of life. It has made outstanding contributions to the global fight against the epidemic. However, a chest CT scan is not a clinical first-line tool. It requires more time to perform and is more expensive, and it is not easy to obtain in remote areas, thus limiting its scope of application. As a first-line clinical tool, diagnosis by chest X-ray (CXR) is the most commonly used method for screening, triage, and diagnosis of various pneumonia (including bacterial, viral, and other types of pneumonia) worldwide. The most common international lung diseases are diagnosed through CXR because it has a faster turnaround time and is more convenient to use in an intensive care environment. On April 15th, Professor Zhang Kang from the Faculty of Medicine of Macau University of Science and Technology, as the main corresponding author, participated in and united multiple units to achieve new achievements in the research of artificial intelligence diagnosis of new coronary pneumonia, and published an article "A deep" in the top international journal "Nature Biomedical Engineering". -learning pipeline for the diagnosis and discrimination of viral, non-viral, and COVID-19 pneumonia from chest X-ray images" (deep learning system for diagnosis and discrimination of viral, non-viral and new coronary pneumonia through CXR imaging).
Huo Wenxun, Dean of the Faculty of Medicine of the University of Science and Technology of China, said that the research team led by Professor Zhang Kang of the Faculty of Medicine has cooperated with multiple units at home and abroad to break through the limitations of multiple sources and heterogeneous data and overcome numerous difficulties. Based on a large number of multi-source heterogeneous CXR data sets, a deep learning system for the diagnosis of new coronary pneumonia and other common lung diseases has been developed, indicating that radiologists can accurately and quickly distinguish between new coronary pneumonia and other types of pneumonia, common lung diseases, and Normal patients and assess their severity. The deep learning system has been tested retrospectively and prospectively on CRX images from four patient cohorts and multiple countries, with an accuracy rate of over 90%. Encouragingly, in a set of independent 440 CXR image diagnostic tests, the system is comparable to the performance of highly qualified radiologists and can also improve the diagnostic accuracy of junior radiologists.
Professor Zhang Kang said that the current lack of a gold standard for clinical evaluation of medical images, the insufficient generalization ability of artificial intelligence systems to be applied to other environments, and the internal decision-making process of deep learning algorithms are still opaque. The clinical transformation and application of artificial intelligence system. To standardize CXR. images, visualize lesions, and accurately diagnose new coronary pneumonia and other common lung diseases, based on 145,000 CXR. images of 120,000 patients, the team successfully developed an accurate artificial intelligence system that marks lesions. The modular processing process including detection, detection, registration, segmentation, and diagnosis and prediction provides stable and interpretable results, which can help radiologists accurately and quickly distinguish between new coronary pneumonia and other types of pneumonia, other common lung diseases, and other common lung diseases. Normal patients and assess their severity, and can also identify the advantages of diffuse lung lesions that are difficult to be detected by radiologists.
The automated deep learning system can be used for the assessment of pneumonia and common lung diseases, and can be used as a first-line diagnostic tool in emergency departments, remote areas or developing countries, promote early intervention, and provide important support for clinical decision-making. Because the system can be quickly deployed to medical centers, providing first-line assessments, and quick turnaround time, it is essential to solve public health problems. This technology will further advance precision medicine and smart medicine, and make an important contribution to global anti-epidemic work. A large number of CXR. image data and codes in this study are open to the world for anti-epidemic, and released to the world by the Greater Bay Area node of China National Bioinformatics Center.
For research papers, please visit the link of "Nature do engineering": https://www.nature.com/articles/s41551-021-00704-1
Comments
Post a Comment