//
πŸ₯
NUS-MIT Healthcare AI Datathon 2020
Search
Try Notion
πŸ₯πŸ₯
NUS-MIT Healthcare AI Datathon 2020
Title: Domain Shift Problem in Chest X-ray Diagnosis AI Model
I have led the ROSE Lab Team winning 2nd Runner-up in NUS-MIT Healthcare AI Datathon 2020
A team from the ROSE Lab took the 2nd Runner-up position among 50 teams from all over the world at the 2020 NUS-MIT Healthcare AI Datathon 2020. Our team was led by the research fellow, Dr. Lin Shan, with project officer Rahul Ahuja and 3 Ph.D. candidates, Yang Siyuan, Wang Yufei, and Li Ling. The team was supervised by Prof Alex Kot, the Director of the ROSE Lab at NTU. This three-day datathon, on 11-13 December 2020, was organized by the National University of Singapore (NUS), National University Health System (NUHS), and MIT Critical Data. Its aim is to bring together clinicians, data scientists and innovators to address current problems in healthcare with data analytics technologies.
Our team focused on solving the domain shift problem in deep learning chest x-ray diagnosis AI models. In the datathon, our team demonstrates the cross-dataset performance degradation of the models trained from Chexpert, ChestX-Ray8, and MIMIC-CXR-JPG datasets. Our team managed to mitigate the problem by using the classic domain adaptation method: MMD and our domain generalization method LDDG [1].
[1] H. Li, Y. Wang, R. Wan, S. Wang, T.-Q. Li, and A. C. Kot, β€œDomain Generalization for Medical Imaging Classification with Linear-Dependency Regularization,” in Proc. Advances in Neural Information Processing Systems (NeurIPS), 2020, no. NeurIPS.