On June 25, 2021, the online workshop titled “AI in Medical Imaging: How to develop models and standardize datasets” was organized by Vingroup Big Data Institute, attracting nearly 7500 viewers on Facebook.
Within the framework of VinDr’s one-year anniversary, Vingroup Big Data Institute organized a workshop “AI in Medical Imaging: How to develop models and standardize datasets”. Not only taking a look back at the achievements of VinDr ecosystem, the event also aims to share with the community some of the most up-to-date knowledge about AI trends in healthcare sector, as well as a variety of techniques in optimizing deep learning models and processing large-scale datasets.
With the above purposes, at the workshop, there is presentation of experienced experts and engineers from the Medical Imaging Center (VinBigdata), who were directly involved in developing VinDr solutions. Specifically, from a research perspective, PhD. Pham Huy Hieu (PhD of Toulouse University, France) gave an overview process for developing AI in diagnostic imaging. Following, engineers Nguyen Ba Dung and Nguyen Trung Nghia respectively shared their experience in algorithm building, data processing and how to approach open standardized datasets.
According to Dr. Pham Huy Hieu, medical imaging is now one of the fields that have the greatest potential for AI application. AI solutions can assist radiologists in detecting abnormalities at early stages and with high accuracy, creating a premise for effective treatment. However, some of the main challenges for the AI development include the lack of standardized and labeled datasets, limited explainability of AI, etc. Therefore, at VinBigdata, to step by step go through those challenges, Center of Medical Imaging implement 03 stages, including collecting and normalizing large-scale datasets; developing and optimizing AI algorithms; evaluation and clinical trials.
For the stage of algorithm development and data processing, engineer Nguyen Ba Dung outlined a specific route, going from problem identification, standardization of labeled data, noise filtering, to operations, testing, optimization, model evaluation and data addition for retraining. In particular, from the presentation of engineer Ba Dung, the community got some suggestions on suitable deployment models for high efficiency such as EfficientNet, ResNet, etc.
In addition to up-to-date knowledge on AI trends and applications in medical imaging, the workshop also provided necessary skills for accessing open datasets and annotation tools on the Internet. Through the visualized introduction, engineer Nguyen Trung Nghia gave the viewers an opportunity to learn the usage of chest X-ray, spine X-ray and rib image data shared via https://lab.vindr.ai/. Moreover, he also presented how to label, manage data and customize source code on the VinDr Lab platform to serve the specific purposes of organizations and individuals . This is expected to be an effective solution to help the Vietnamese AI community overcome the lack of large-scale, standardized and labeled medical datasets.
During 2 hours of the event, the viewers not only listened to the experts and engineers’ sharing on building AI solutions, but also had the opportunity to interact directly, thus being given answers to relevant questions. With practical content, the seminar attracted nearly 7500 online followers, 166 shares and more than 800 responses on the event page.
This is one of our comprehensive efforts to promote AI development in Vietnam. In addition to researching and building practical solutions, meanwhile sharing healthcare datasets and tools for the community, the development team also hopes to inspire young talents for improving their technology capacity, thereby creating a bright future for Vietnam’s science.
The event was livestreamed on VinBigdata’s fanpage. If interest, you can follow here.