A new study published in Cancer Imaging introduces DynamicFusionNet, a decoder framework that dynamically decides, for each input, how features from different layers, scales and representational views should be merged, motivated by the observation that standard segmentation architectures rely on static fusion that often fails to accommodate the heterogeneity of medical images. Described as a drop-in decoder that learns dense fusion policies conditioned on the input, the framework has three main components: pixel-wise gating on skip pathways, allowing selection at the level of individual pixels of which layer features to pass forward; a multi-scale calibration module that integrates complementary branch responses through a shared pointwise projection; and a dual-pathway global-local arbitration design that balances high-level semantics with fine boundary details. The modules can be integrated into standard encoder-decoder pipelines without modifying the backbone, and the authors position DynamicFusionNet as a modality-agnostic fusion principle adoptable in existing segmentation systems. The study reports experiments on five public benchmarks, with main results on BraTS 2021, RETOUCH and PARSE 2022, covering brain tumour imaging with multiple magnetic resonance modalities, retinal imaging, and cardiac structures. According to the abstract, DynamicFusionNet consistently outperforms strong CNN and Transformer baselines in segmentation accuracy and boundary fidelity across these settings.