The Importance of Background Information for Out of Distribution Generalization
This addresses domain generalization for trustworthy machine learning in healthcare, but it is incremental as it builds on existing methods with task-specific masks and data scaling.
The paper tackles the problem of domain generalization in medical image classification by showing that background regions, not just abnormality segmentations, provide helpful signal for out-of-distribution performance. The result is that using task-specific masks covering all relevant regions significantly improves existing methods, with better generalization than empirical risk minimization requiring scaled-up training data.
Domain generalization in medical image classification is an important problem for trustworthy machine learning to be deployed in healthcare. We find that existing approaches for domain generalization which utilize ground-truth abnormality segmentations to control feature attributions have poor out-of-distribution (OOD) performance relative to the standard baseline of empirical risk minimization (ERM). We investigate what regions of an image are important for medical image classification and show that parts of the background, that which is not contained in the abnormality segmentation, provides helpful signal. We then develop a new task-specific mask which covers all relevant regions. Utilizing this new segmentation mask significantly improves the performance of the existing methods on the OOD test sets. To obtain better generalization results than ERM, we find it necessary to scale up the training data size in addition to the usage of these task-specific masks.