Dominance guided method improves medical severity grading from sparse comparisons
DOSE: Dominance-Guided Ordinal Severity Estimation from Sparse Relative Annotations
Computer Vision and Pattern Recognition
Summary
Medical images are often rated using fixed categories like mild or severe, but these categories have a natural order that many methods ignore. The authors propose a method called DOSE that learns how severe a medical condition is by comparing pairs of images instead of needing exact labels. DOSE creates a global ranking from these pairwise comparisons and uses this to assign scores, which it then converts to the usual categories without needing labeled examples. This approach worked better than previous methods on multiple medical image datasets.
What this means in practice
- •For clinical imaging teams: Estimate medical condition severity from fewer annotated images by using sparse pairwise severity comparisons instead of absolute labels.
- •For medical ai developers: Build models that provide continuous severity scores and convert them into standard discrete categories without needing large labeled datasets.$Commercial implications: Enables development of diagnostic tools with improved severity estimation for sale to healthcare providers.
Authors
Tushir Sahu, Hideaki Hayashi, Hajime Nagahara
Abstract
Medical severity estimation commonly relies on absolute grade annotations, which are fundamentally ordinal, yet existing methods often treat the grades as independent categories. Learning-to-rank methods, which can be trained using relative annotations based on severity comparisons of image pairs, can provide continuous severity rankings but cannot directly predict discrete severity grades. We propose Dominance-guided Ordinal Severity Estimation from Sparse Relative Annotations (DOSE), a framework for learning ordinal medical severity from sparse pairwise comparisons without using absolute annotations. DOSE aggregates the observed comparisons through a global comparison graph to estimate a dominance score for each image. Differences between dominance scores set cumulative margins in the embedding space, so that pairs with larger estimated severity differences are separated more strongly, and also generate pseudo-pairs that supervise a scalar severity head alongside the observed comparisons. The learned continuous severity axis is converted into discrete grades via label-free score calibration, with tie-anchored segmentation as the main calibration method. In the experiment, DOSE consistently outperformed pairwise-ranking baselines on ACNE04, IDRiD, and LIMUC under different pair budgets. Code is available at https://github.com/TushirSahu/DOSE