Release date: 2018-02-23
Artificial intelligence is a hot spot in recent years. With AlphaGo defeating Li Shizhen as a symbol, artificial intelligence quickly entered the field of ordinary people. Along with it, it is another breakthrough in the field of medical health. Previously, we reported an artificial intelligence that could diagnose breast cancer. It takes only a few seconds to get an accurate diagnosis than a human doctor spends dozens of hours. This also officially declares that artificial intelligence has surpassed humans in medical image-based diagnosis, and the gap will only widen.
â–² Professor Zhang Kang, the author of this study (Source: UCSD)
The study, which is on the cover of Cell, today provides us with a new AI tool. Professor Zhang Kang of the study is a professor of ophthalmology at the University of California, San Diego (UCSD) and is the chief physician of ophthalmology genetics (Chief, Ophthalmic Genetics). In ophthalmic treatment, retinal OCT (optical coherence tomography) imaging technology is one of the most commonly used diagnostic techniques, with more than 30 million uses per year. By taking high-resolution images of retinal tissue, doctors can accurately diagnose and provide treatment options for blinding eye diseases such as age-related macular degeneration and diabetic macular edema.
Based on the universality of OCT technology, if we can use the rapidly developing AI technology to process these images, it will undoubtedly further improve the efficiency of diagnosis and even improve the accuracy of diagnosis. To this end, Professor Zhang Kang obtained more than 200,000 images of OCT and trained a deep learning algorithm using 100,000 images from nearly 5,000 patients. After a lot of iterative training, the accuracy of this algorithm has reached a peak.
â–²The design process of the research (Source: "Cell")
The researchers also used choroidal neovascularization, diabetic macular edema, drusen, and normal retinal OCT images to test the algorithm. The study found that the overall accuracy of the AI ​​tool reached 96.6%, the sensitivity was 97.8%, the specificity was 97.4%, and the AUC value (which can reflect the pros and cons of the algorithm) was as high as 99.9%.
Subsequently, the researchers looked for six experts with extensive clinical experience to compare their diagnostic results with the diagnostic results of the AI ​​tool. Studies have shown that there is no significant difference in specificity and sensitivity. In other words, we can safely let this AI tool make a diagnosis. The characteristics of AI tools are able to achieve large-throughput screening that humans cannot.
â–²This tool can accurately diagnose a variety of retinal abnormalities, and there is no significant difference with the results of human experts (Source: "Cell")
"Macular degeneration and diabetic macular edema are two major irreversible causes of blindness, but as long as they are discovered early, they can be treated," Professor Zhang Kang said. "In the past, only a few experts were able to decide how to treat and when to treat. They need to be trained for many years and often focus on the city. Our AI tools can be used anywhere in the world, especially in remote areas. This is especially important in areas with relatively low medical resources such as China, India, and Africa. ."
What is even more gratifying is that Professor Zhang Kang’s team has shown that this AI tool has a wide applicability. They use the same deep learning framework to make accurate diagnoses of childhood pneumonia. According to estimates by the World Health Organization, pneumonia kills about 2 million children under the age of five each year and is one of the leading causes of child death. Children's pneumonia can be divided into bacterial and viral depending on the pathogen, and the treatment of the two pneumonias is different - the former requires antibiotic treatment and the latter requires other treatments. Therefore, the timely differentiation of these two types of pneumonia is essential for the treatment of children.
â–²This AI tool can also distinguish between bacterial and viral pneumonia in children (Source: Cell)
Similarly, the researchers collected 5,232 chest X-rays for training in the AI ​​system. After iteration and testing, this AI tool for the diagnosis of childhood pneumonia achieved 92.8% accuracy, 93.2% sensitivity, 90.1% specificity, and 96.8% AUC. These data indicate that AI is sufficient to distinguish between bacterial and viral pneumonia.
“If we can work closely together, we can develop better and better diagnostic techniques with increasing computing power,†added Professor Zhang Kang. “In the future, we will have more data, more computing power, and More experience from people using this system. We can control costs and bring the best possible treatment to patients."
We once again congratulate Professor Zhang Kang on this breakthrough, and I hope that artificial intelligence can provide more accurate and efficient diagnosis in the future and save the lives of patients.
Reference materials:
[1] Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning
[2] Artificial intelligence can diagnose and triage retinal diseases
Source: Academic Jingwei
3M Vhb Foam Tape,Gray Acrylic Tape,Acrylic Tape Adhesive,Double Sided Acrylic Foam Tape
Kunshan Jieyudeng Intelligent Technology Co., Ltd. , https://www.jerrytapes.com