The Medical Imaging Department, the Institute of Big Data (under Vingroup), has successfully built a test version of the software that automatically interprets chest X-ray images and supports disease diagnosis.
At the Pneumothorax Segmentation diagnostic competition organized by the American Society for Imaging Informatics in Medicine (SIIM), the Institute is in the top 5 out of a total of more than 1000 contestant teams from all over the world.
The software diagnoses diseases with accuracy up to 90%
The disease diagnosis via medical images (such as X-ray, CT, MRI) is often performed manually by radiologists. This approach has certain shortcomings in terms of accuracy. Doctors can sometimes miss small but important details due to visual limitations, time pressure, number of patients, as well as work intensity.
This limitation will be overcome when applying artificial intelligence (AI) to image diagnosis. The approach for chest radiography has been successfully implemented by the Medical Imaging team of the Institute of Big Data (under Vingroup). This is a type of routine diagnosis and remains in great demand, accounting for 70-80% of diagnostic imaging in hospitals in Vietnam as well as in the world.
Professor Vu Ha Van, Scientific Director of the Institute of Big Data, said: “In order to ensure the competency to conduct international-standard research, the Medical Imaging team has sent the software to 2 prestigious competitions held on an open platform. Specifically, in the competition to diagnose pneumothorax (Pneumothorax Segmentation) organized by the American Society for Imaging Informatics in Medicine (SIIM), the Institute’s research team is in the top 5 in total of more than 1000 teams from all over the world, including many prestigious organizations of the Russian Federation, the United States, China, Israel, etc.
Additionally, in the competition to diagnose 12 common lung diseases via chest radiographs (CheXpert) organized by Stanford University – USA, the team ranked 5th out of 40 teams. In particular, on a specific set of radiographs, the algorithm’s diagnostic results were better or equivalent compared to the panel of radiologists organized by Stanford to evaluate the competition.”
According to Dr. Nguyen Quy Ha, Head of Medical Imaging (the Institute of Big Data), if based on the evaluation criteria of the medical industry, both the sensitivity and specificity of the algorithm were over 90%, even over 95% for some diseases. About processing speed, it would take 5-10 minutes to read a standard X-ray image, but the algorithm only took 2-5 seconds depending on the hardware configuration.
Mr. Ha added that although initial positive results have been achieved, the algorithm was being built based on foreign open data sources. If it is possible to collect medical image data of Vietnamese people, the accuracy and applicability in practice in Vietnam would be really high.
Input data needed
Dr. Nguyen Quy Ha also shared the current difficulty was that standard medical dataset cannot be developed in days. For each type of disease, the algorithm needs to take images from about 100,000 to 200,000 patients, which is not a small number. “The bigger the input, the better the algorithm. Just like doctors, the more diverse cases a doctor is exposed to, the more experience they have and the more accurate the diagnosis,” said Dr. Ha.
On July 23, 2019, at the conference “Promoting the implementation of electronic medical records towards hospitals that do not use paper medical records and do not use cash to pay for hospital fees”, organized by the Ministry of Health in Da Nang, the Institute of Big Data signed a Memorandum of Understanding with the Information Technology Department – Ministry of Health. The two sides agreed to cooperate in developing regulations on the exchange of medical examination and treatment data between organizations and individuals for the research and development of AI products in Health.
“After being standardized and fully annotated, the data will not carry any personal information and will be shared widely with the community. When the software is complete, hospitals can be used for free. Doctors will receive maximum support, minimizing time to interpret images. In lower-level hospitals, where there are not many specialists, the application of AI will significantly increase the opportunity for patients to have access to modern diagnostic methods, helping radiologists have more useful information to decide on therapy”, Prof. Vu Ha Van said.
When the product is completed, one possible option is to deploy a remote diagnostic system that applies cloud computing technology. Users will be granted an account and just need to send images to have reading results sent back soon after. This approach has the advantage of being economical and hospitals do not need to invest in their own infrastructure.
Prof. Vu Ha Van also shared: “To put the software into use, there is still a lot of work to be done. In addition to collecting data of Vietnamese people, there are also legal issues and licensing procedures. We strive to bring the software meet international standards and become useful to hospitals in the country. We will also aim for the necessary diagnosis for some dangerous and common diseases in Vietnam such as cancers or diseases of the heart, nerves and diabetes.”