Regression Metric Loss: Learning a Semantic Representation Space for Medical Images

AbstractRegression plays an essential role in many medical imaging applications for estimating various clinical risk or measurement scores. While training strategies and loss functions have been studied the deep neural networks image classification tasks, options regression tasks are very limited. One of key challenges is that high-dimensional feature representation learned by existing popular like Mean Squared Error L1 hard to interpret. In this paper, we propose a novel Regression Metric Loss (RM-Loss), which endows space with semantic meaning label finding manifold isometric space. Experiments on two i.e. coronary artery calcium score estimation bone age assessment, show RM-Loss superior losses both performance interpretability. Code available at https://github.com/DIAL-RPI/Regression-Metric-Loss.KeywordsMedical regressionMetric learningRepresentation learningSemantic representationInterpretability