{"id":792212,"date":"2021-11-16T08:00:27","date_gmt":"2021-11-16T16:00:27","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&p=792212"},"modified":"2021-11-17T09:17:59","modified_gmt":"2021-11-17T17:17:59","slug":"research-talk-causality-for-medical-image-analysis","status":"publish","type":"msr-video","link":"https:\/\/www.microsoft.com\/en-us\/research\/video\/research-talk-causality-for-medical-image-analysis\/","title":{"rendered":"Research talk: Causality for medical image analysis"},"content":{"rendered":"
Machine learning has huge potential to augment medical image analysis workflows and improve patient care. However, two of its notorious real-world challenges are the difficulty in acquiring sufficient, high-quality annotated data and mismatches between the development dataset and the target environment (across hospitals, for example).<\/p>\n
Daniel Coelho de Castro, a researcher in the Health Intelligence Group at Microsoft Research Cambridge, will discuss how causal reasoning can shed new light on these pervasive issues and appropriate mitigation strategies. In particular, a causal perspective enables decisions about data collection, annotation, pre-processing, and learning strategies to be made\u2014and scrutinized\u2014more transparently. He will highlight how understanding and communicating the story behind the data helps improve the reliability of machine learning systems in high-risk healthcare settings. This session will cover a causal categorization of potential biases when developing medical imaging models, a couple of worked clinical examples, and step-by-step recommendations for practitioners.<\/p>\n