基于卷积神经网络胃癌分割与T分期算法

基于胃癌CT图像准确分割胃癌和精准预测胃壁肿瘤浸润深度对于筛查胃部疾病、临床诊断、术前预测、术后评估计划至关重要。为了准确地从胃癌CT图像分割出胃癌并对肿瘤进行定性分期，提出一种基于卷积神经网络的胃癌分割与T分期算法（SC-Net）。SC-Net有两条主干线：分割主线、分类主线。这种新型算法分为两步进行训练：第一步只训练分割主线得到肿瘤的粗分割结果，然后在第一步基础之上联合训练分割分类主线得到最终的精分割和肿瘤T分期结果。为了提高算法对胃癌区域的关注度，提出了注意力机制加强算法的准确性。此外还使用多核残差模块和密集连接空洞卷积模块提取深层的特征信息。对所提算法进行定性定量分析。实验表明所提方法在胃癌分割和T分期上均优于同类方法，所提方法有作为筛查胃部疾病、辅助医生诊断的潜力。

Gastric cancer segmentation and T staging algorithm based on convolutional neural network

The accurate segmentation of gastric cancer and the accurate prediction of gastric wall tumor invasion depth based on gastric cancer CT images are essential for screening of gastric diseases,clinical diagnosis,preoperative prediction and postoperative evaluation planning.A gastric cancer segmentation and T staging algorithm based on convolutional neural network(SC-Net)is proposed to accurately segment gastric cancer in CT images and to qualitatively stage the tumors.SCNet has two main lines,namely segmentation main line and classification main line.For the novel algorithm,the training is divided into two steps.The first step is to train only the main line of segmentation for obtaining the rough segmentation result,and then on the basis of the first step,the main lines of segmentation and classification are jointly trained to obtain the final fine segmentation and T staging results.In order to improve the algorithm's attention to the gastric cancer region,an attention mechanism is proposed to enhance the accuracy of the algorithm.In addition,multi-core residual module and densely connected dilated convolution module are used to extract underlying characteristic information.The qualitative and quantitative analyses on the proposed algorithm show that the proposed method is superior to similar methods in gastric cancer segmentation and T staging.The proposed method has the potential to screen stomach diseases and assist doctors in diagnosis.