Vision-based artificial intelligence recognition technology: Identify lung cancer and breast cancer in as little as 30 seconds
Release date: 2017-09-12 Source: Arterial Network Shower Gel,Shower Cream,Body Shower Gel,Foaming Shower Gel Guangzhou Lingxue Cosmetics Co., Ltd , https://www.gzlxgj188.com
Wang Pingan (left) and the research team successfully applied artificial intelligence image recognition technology to improve the accuracy of image diagnosis of lung cancer and breast cancer.
Complete identification for lung cancer and breast cancer in 30 seconds
Lung cancer and breast cancer are common high-risk ailments in Hong Kong. To improve the efficiency of clinical diagnosis, the research team of the Chinese University of Hong Kong Engineering Institute uses artificial intelligence image recognition technology to interpret medical tomography and pathological tissue slices through deep learning systems to study the disease. Two types of cancer, the results show that the accuracy of cancer medical images interpreted by this technology is as high as 91% and 99%, respectively, and the recognition process is accelerated to 30 seconds to 5 minutes, which reduces the rate of misdiagnosis. The team will work with local public hospitals and expect the technology to be widely available in the next year or two.
Lung cancer is the number one fatal cancer in Hong Kong, with thousands of new cases every year. Most of the early stage of lung cancer occurs in the form of small pulmonary nodule, and the patient's lung image shows a small group shadow. At present, doctors mainly through chest computed tomography (CT) image examination to see if there is a small pulmonary nodule in the patient's lungs.
It takes five minutes to see the naked eye.
However, each inspection can produce up to hundreds of CT images. In general, if you observe them one by one with a naked eye, each flower takes 3 seconds and takes at least 5 minutes. It takes a lot of time and effort, and the accuracy is also because There are differences in the doctor's experience and mental state.
Wang Pingan, a professor at the Department of Computer Science and Engineering at CUHK, and his team started the experiment five years ago. Using Deep Learning to interpret CT scan images, they automatically identify locations where small pulmonary nodules may occur in just 30 seconds, with an accuracy of 91%.
Wang Ping explained that deep learning means that the computer imitates the human brain. According to the collected data, according to the doctor's instructions and demonstrations, the data analysis, and repeated learning and modification, can improve the system accuracy.
It is expected to be widely used within one or two years.
He believes that the technology will be widely used in the next year or two. Deep learning improves the sensitivity of technology and reduces the false positive rate through advanced methods, solving the biggest challenge of detecting images with the naked eye.
Wang Pingan also revealed that the team will jointly develop products related to the three hospitals in Beijing, and will cooperate with Hong Kong hospitals to apply them in the local medical system as soon as possible.
In addition to lung cancer, the technology can also be applied to the diagnosis of breast cancer. Doctors usually use mammography or MR scanning to detect the position of the lumps. When detecting lymph node metastasis, a small piece of living tissue is taken as a sample, the lymph nodes are examined under the microscope for metastasis, and the tumor is judged to be benign or malignant.
The resolution of a digital bio-sliced ​​full-slice image is very high, and the file size can reach 1GB. So the research team developed a new deep-layered convolutional neural network to process the sliced ​​images of breast cancer in stages.
First, use the improved Fully Convolutional Network, a fast predictive model that performs coarser but more sensitive images, reconstructs more accurate and accurate predictions, and finally locates and selects lymph nodes. Transferred image.
The entire process takes only about 5 minutes to 10 minutes, and currently it takes 15 minutes to 30 minutes for a visual inspection alone. The accuracy of automated testing is approximately 99%.
Dou Qi, a Ph.D. student member of the research team, pointed out that the team had participated in many international academic competitions earlier. The competition also provided patient data to test the accuracy of the system. The results showed that the team of the Zhongda team performed well in detecting lung cancer and breast. The accuracy of cancer is 90% or more.
>>>> Character introduction
Wang Pingan: Visionary Medical Co-founder & Honorary Chairman, Education Minister Jiang Scholar, Professor of Computer Science and Engineering, The Chinese University of Hong Kong, Director of Virtual Reality, Visualization and Imaging Research Center, Director of Computer Research in Key Strategic Research, Chinese Academy of Sciences Director of Human-Computer Interaction Laboratory of Shenzhen Advanced Institute. His main research interests include computer-aided medicine, virtual reality in medicine, interactive scientific computing visualization, and three-dimensional medical images. He has published more than 400 academic papers. The first "virtual person" was led by Professor Wang Pingan, and for the first time on the computer platform, the highly interactive and realistic visualization of ultra-high resolution visible human data was realized.