Imbalance-Aware Self-Supervised Learning for 3D Radiomic Representations
This addresses data imbalance issues in medical imaging for radiomic analysis, though it appears incremental as it builds on existing self-supervised and imbalance-handling techniques.
The paper tackles the challenge of learning 3D medical image representations under data imbalance by proposing a self-supervised framework using a 3D Siamese network with unsupervised strategies like sample re-weighting and balanced batch composition. The result shows significant improvement in brain tumor classification and lung cancer staging tasks when combining learned features with traditional radiomics.
Radiomic representations can quantify properties of regions of interest in medical image data. Classically, they account for pre-defined statistics of shape, texture, and other low-level image features. Alternatively, deep learning-based representations are derived from supervised learning but require expensive annotations from experts and often suffer from overfitting and data imbalance issues. In this work, we address the challenge of learning representations of 3D medical images for an effective quantification under data imbalance. We propose a \emph{self-supervised} representation learning framework to learn high-level features of 3D volumes as a complement to existing radiomics features. Specifically, we demonstrate how to learn image representations in a self-supervised fashion using a 3D Siamese network. More importantly, we deal with data imbalance by exploiting two unsupervised strategies: a) sample re-weighting, and b) balancing the composition of training batches. When combining our learned self-supervised feature with traditional radiomics, we show significant improvement in brain tumor classification and lung cancer staging tasks covering MRI and CT imaging modalities.