Class Incremental Medical Image Segmentation via Prototype-Guided Calibration and Dual-Aligned Distillation

The recent submission titled 'Class Incremental Medical Image Segmentation via Prototype-Guided Calibration and Dual-Aligned Distillation' introduces two novel methods aimed at addressing challenges in Class Incremental Medical Image Segmentation (CIMIS). Traditional approaches often fail to maintain the integrity of previously learned classes while incorporating new data, leading to knowledge degradation. The proposed PGCD method enhances the preservation of reliable old knowledge by calibrating class-specific distillation intensity based on prototype-to-feature similarity. Meanwhile, DAPD aligns local prototypes of old classes with both global and local prototypes, significantly boosting segmentation performance. Evaluations on widely used multi-organ segmentation benchmarks demonstrate that these methods outperform existing state-of-the-art techniques, showcasing their robustness and generalization capabilities. This advancement is crucial for improving medical imaging applications,…