Abstract
We propose a novel multi-stage trans-dimensional architecture for multi-viewcardiac image segmentation. Our method exploits the relationship betweenlong-axis (2D) and short-axis (3D) magnetic resonance (MR) images to perform asequential 3D-to-2D-to-3D segmentation, segmenting the long-axis and short-axisimages. In the first stage, 3D segmentation is performed using the short-axisimage, and the prediction is transformed to the long-axis view and used as asegmentation prior in the next stage. In the second step, the heart region islocalized and cropped around the segmentation prior using a Heart Localizationand Cropping (HLC) module, focusing the subsequent model on the heart region ofthe image, where a 2D segmentation is performed. Similarly, we transform thelong-axis prediction to the short-axis view, localize and crop the heart regionand again perform a 3D segmentation to refine the initial short-axissegmentation. We evaluate our proposed method on the Multi-Disease, Multi-View& Multi-Center Right Ventricular Segmentation in Cardiac MRI (M&Ms-2) dataset,where our method outperforms state-of-the-art methods in segmenting cardiacregions of interest in both short-axis and long-axis images. The pre-trainedmodels, source code, and implementation details will be publicly available.