全局与局部注意力机制的胃部肿瘤分割算法

Gastric tumor segmentation algorithm based on global-local attention mechanism

通过CT实现术前胃部肿瘤诊断是一种潜在高效的技术方法,而准确的肿瘤影像分割是实现该方法的关键。为了能够精确地提取到肿瘤区域,提出一种基于注意力机制的2D分割网络GLat-Net对上腹部CT影像中的胃部肿瘤区域进行分割,通过增加对肿瘤周围区域的关注,从全局和局部两个角度提取有效的上下文信息;同时在解码模块中引入权重模块突出具有代表性的特征。通过实验结果证明,相比较于其他前沿分割方法,该算法在胃部肿瘤分割上有更高的准确度。

Using CT to realize preoperative gastric tumor diagnosis is a potentially efficient technical method,of which accurate tumor image segmentation is the prerequisite for preoperative diagnosis.In order to achieve a better performance on the extraction of tumor,a 2D segmentation network(GLat-Net)based on attention mechanism is proposed to complete the segmentation of the gastric tumor on upper abdominal CT images,which focuses on the area around the tumor and extracts effective contextual information from global and local perspectives.Furthermore,the weight module is introduced to decoder module for highlighting the representative features.The experiment results show that the proposed approach outperforms current state-of-the-art methods on gastric tumor segmentation,having a higher segmentation accuracy.