Dilated convolution network with edge fusion block and directional feature maps for cardiac MRI segmentation

• DOI: 10.3389/fphys.2023.1027076
• Publication Date: 2023-01-26T08:23:50Z
• Authors: Zhensen Chen; Jieyun Bai; Yaosheng Lu
• Keywords: Feature (linguistics); Upsampling; Convolution (computer science)

Cardiac magnetic resonance imaging (MRI) segmentation task refers to the accurate of ventricle and myocardium, which is a prerequisite for evaluating soundness cardiac function. With development deep learning in medical imaging, more heart methods based on have been proposed. Due fuzzy boundary uneven intensity distribution MRI, some existing do not make full use multi-scale characteristic information problem ambiguity between classes. In this paper, we propose dilated convolution network with edge fusion block directional feature maps MRI segmentation. The uses module preserve information, adopts direction field obtain improve original features. Firstly, obtained fused through convolutional layers different scales while downsampling. Secondly, decoding stage, integrates features into side output encoder concatenates them upsampled Finally, concatenated utilize generate final result. Our method conducts comprehensive comparative experiments automated diagnosis challenge (ACDC) myocardial pathological (MyoPS) datasets. results show that proposed has better performance compared other methods.