基于多视角融合的医学图像分割方法研究

A Method of Medical Image Segmentation Based on Multi-view Fusion

提出了一种基于变换器的多视角医学图像融合分割框架,通过采用一种发散融合的注意力模型来建模跨视图信息,充分挖掘语义之间的相互依赖,解决了不同视图之间的数据不对齐以及难以捕捉长距离相关性等关键问题。同时使用多尺度注意力机制来提取多尺度的全局特征。在Cardiac MRI数据集上分别评估了多疾病、多视角和多中心的右心室分割任务,验证了该方法具有更优的分割性能。

A segmentation framework of medical image fusion by multi-view based on a transformer is proposed,which uses a divergent fusion attention model to establish the cross view information model,fully mines semantic interdependence,solves key issues such as data misalignment between different views and difficulty in capturing long-distance correlations,and uses multi-scale attention mechanisms to extract global features at multiple scales at the same time.It evaluates the right ventricular segmentation tasks of multiple diseases,multiple perspectives and multiple centers on the Cardiac MRI data set,and verifies that the method has better segmentation performance.