2D medical image segmentation via learning multi-scale contextual dependencies

•We propose a novel medical image segmentation framework based on deep learning.•This work learns multi-scale contextual dependencies from two dimensions.•Our framework significantly reduces the missed diagnosis rate compared to others.•This work successfully avoids feature disappearance of small ROIs in deep layers.•Visualization comparisons of learned features boost networks’ interpretability.•This work conducts comprehensive experiments on three real-world medical datasets.