Vector Researcher Bo Wang Develops Method for Harmonizing Multi-Hospital Medical Data
Vector Institute faculty member Bo Wang has developed a novel method called Moment Matching for Multi-source Domain Adaptation (M3SDA) to address the challenge of harmonizing medical data from multiple hospitals. Traditional machine learning models often struggle when applied across different institutions due to variations in data collection and sorting methods. Wang’s approach enables the integration of data from diverse sources, allowing models trained on one hospital's data to be effectively applied to another. Published in the Proceedings of the IEEE International Conference on Computer Vision, the M3SDA model creates a common feature space and matches distributions across domains. It has already generated a dataset of 600,000 annotated images across 345 categories. The method is currently being utilized by researcher Lauren Erdman at SickKids Hospital for studies involving ultrasound data harmonization and uroflow analysis. Wang highlights that this technique simultaneously addresses the scarcity of annotated medical images and the lack of generalizability in existing models, offering significant potential for improving medical image analysis and AI-driven healthcare solutions.
Wire timeline
Vector Researcher Bo Wang Develops Method for Harmonizing Multi-Hospital Medical Data
Vector Institute faculty member Bo Wang has developed a novel method called Moment Matching for Multi-source Domain Adaptation (M3SDA) to address the challenge of harmonizing medical data from multiple hospitals. Traditional machine learning models often struggle when applied across different institutions due to variations in data collection and sorting methods. Wang’s approach enables the integration of data from diverse sources, allowing models trained on one hospital's data to be effectively applied to another. Published in the Proceedings of the IEEE International Conference on Computer Vision, the M3SDA model creates a common feature space and matches distributions across domains. It has already generated a dataset of 600,000 annotated images across 345 categories. The method is currently being utilized by researcher Lauren Erdman at SickKids Hospital for studies involving ultrasound data harmonization and uroflow analysis. Wang highlights that this technique simultaneously addresses the scarcity of annotated medical images and the lack of generalizability in existing models, offering significant potential for improving medical image analysis and AI-driven healthcare solutions.
Vector Institute for Artificial Intelligence