Only taking a few seconds for each study, VinDr can detect, localize and diagnose a variety of lesions on the lung, liver, breast and spine with an average accuracy of over 90%. Thus, after only a year of operation, the AI - based solution called VinDr, which is developed by Vingroup Big Data Institute, has become an effective “assistant” for radiologists, especially during the current pandemic.
An assistant for radiologists
One year ago, VinBigdata announced to launch VinDr – the first version of a comprehensive artificial intelligence (AI) solution for medical imaging.
During this period, VinDr had 4 diagnostic modules on chest X-ray, mammography, chest CT and liver CT. The clinical trial of the two modules which diagnose lung pathology and breast cancer on X-ray images showed that VinDr had a positive influence on the judgment of radiologists (the implementation took place at Hanoi Medical University Hospital and 108 Military Central Hospital). Specifically, the consensus among doctors increased by 8.7% after referring to the diagnostic results of the AI.
After more than a year of ceaseless efforts, VinBigdata has successfully developed a new module for interpreting spine X-ray images. It can detect 6 types of common lesions in Vietnam, including: osteophytes, disc space narrowing, surgical implant, foraminal stenosis, spondylolisthesis and vertebral collapse, with accuracy rate of 90%.
PhD. Ha Quy Nguyen (Director of Medical Imaging Center, VinBigdata) said that the team was launching a new project to apply AI in gastrointestinal endoscopy. This solution aims to bring AI integrated with the endoscope, which means that when the doctor moves the endoscope probe, AI can immediately find the lesions and give instant suggestions. This is a new direction in diagnostic imaging, promising a step forward for the medical imaging industry of Vietnam.
According to Dr. Ha, along with 5 medical imaging modules that have been released, the two others including brain MRI and brain CT are expected to be completed and put into application by the end of 2021.
“The products put into practice are based on the highest standards of research and development. They are also evaluated by experts from prestigious journals and conferences in the field of medical imaging. This is an important foundation to ensure that the solution will have the best effect and gain the trust of both radiologists and medical professionals.” Dr. Ha added.
Aspire to build a smart healthcare system for Vietnam
Proving both outstanding speed and accuracy, VinDr is currently an effective assistant for radiologists at 11 hospitals and clinics in Hanoi, Phu Tho, Hai Phong, Lam Dong, Binh Dinh, etc. At the same time, the solution also creates favorable conditions for the exchange of diagnostic data between different organizations, aiming to ensure the smooth operation of the healthcare system throughout the country.
Prof. Van Ha Vu, Scientific Director of VinBigdata shared: “Currently, a huge medical force is concentrating for the frontline to fight against Covid-19 pandemic. However, the rate of other serious diseases remained unchanged, increasing unprecedented pressure on medical workforces. Therefore, along with the modules that diagnose lung lesions through chest X-ray and CT images, we also accelerate the development, testing and deployment of AI applications to detect liver, breast and spine diseases. Not only is it an urgent response to the pandemic, we also hope to solve long-term problems for the Vietnamese healthcare sector.”
With the methodical investment of Vingroup, aiming to build a smart healthcare system, contribute to improving the quality of medical examination and treatment for the Vietnamese, VinBigdata not only focuses on developing VinDr solutions, but also strives to share medical data and tools for the whole community.
Only in the first half of 2021, VinBigdata pioneered to open a dataset of 18,000 labeled chest X-ray images through a global competition on building AI models for medical imaging. In addition, VinDr Lab (the first open-source platform for medical image annotation) was also launched by VinBigdata on the Github platform (a service that provides source code repositories). Currently, VinDr Lab has supported to label X-ray images. Other features for tomography and magnetic resonance imaging are under development. They are expected to be shared in the near future as well.
In addition, VinBigdata has been implementing the open medical image data portal. The community can access large-scale datasets via the link: https://lab.vindr.ai. Chest X-rays, SpineX-rays and Rib X-rays are now available there.
Users can view labeled datasets, thereby verifying both quality and reliability of the medical images. At the same time, the VinDr Lab platform also allows users to submit requests to upload medical image data, directly try annotation and data management features. Users can also contribute to improving software quality by giving feedbacks. It aims to standardize the data sharing and spread practical values for the Medical Imaging researchers in particular and the scientific community in general.
The Center of Medical Imaging (Vingroup Big Data Institute) has won many prestigious international awards in recent years such as: No.1 in Abnormal Image Detection in Endoscopy Videos (EndoCV) (ISBI conference 2020); No.1 in Herbarium (FGVC7, CVPR 2020); Top 3 in Pulmonary Embolism Detection Challenge, organized by the Radiological, Society of North America (RSNA), 2020; No.1 in Global Wheat Detection competition (ECCV, 2020); No. 1 in CheXpert contest organized by Stanford University and Top 10 of the brain diagnostic imaging contest organized by RSNA.
