梁礼明, 何安军, 阳渊, 吴健. 基于Swin Transformer和图形推理的结直肠息肉分割方法[J]. 工程科学学报, 2024, 46(5): 897-907. DOI: 10.13374/j.issn2095-9389.2023.04.21.004
引用本文: 梁礼明, 何安军, 阳渊, 吴健. 基于Swin Transformer和图形推理的结直肠息肉分割方法[J]. 工程科学学报, 2024, 46(5): 897-907. DOI: 10.13374/j.issn2095-9389.2023.04.21.004
LIANG Liming, HE Anjun, YANG Yuan, WU Jian. Colorectal polyp segmentation method based on the Swin Transformer and graph reasoning[J]. Chinese Journal of Engineering, 2024, 46(5): 897-907. DOI: 10.13374/j.issn2095-9389.2023.04.21.004
Citation: LIANG Liming, HE Anjun, YANG Yuan, WU Jian. Colorectal polyp segmentation method based on the Swin Transformer and graph reasoning[J]. Chinese Journal of Engineering, 2024, 46(5): 897-907. DOI: 10.13374/j.issn2095-9389.2023.04.21.004

基于Swin Transformer和图形推理的结直肠息肉分割方法

Colorectal polyp segmentation method based on the Swin Transformer and graph reasoning

  • 摘要: 针对结直肠息肉图像分割中病灶区域尺度变化大、边缘模糊以及息肉与正常组织对比度低等问题,导致病变区域分割精度低和分割边界存在伪影,提出一种基于Swin Transformer和图形推理的自适应网络. 该网络一是利用Swin Transformer编码器逐层提取输入图像的全局上下文信息,弱化背景信息干扰,多尺度分析病变区域的显著性特点. 二是提出全局与局部特征交互模块增强网络对复杂病灶的空间感知能力,突出待分割目标的关键位置信息,精准定位目标. 三是通过区域引导图推理模块以图循环递推的方式挖掘先验信息之间的高阶显性关系,促进图间信息传递. 四是设计面向边缘细节的边缘约束图推理模块,整合边缘细节,改善分割效果,提高分割精度. 在CVC-ClinicDB、Kvasir、CVC-ColonDB和ETIS数据集上进行实验,其Dice系数分别为0.939,0.926,0.810和0.788,平均交并比分别为0.889,0.879,0.731和0.710,分割性能优于现有方法. 仿真实验结果表明,对于形态结构复杂、对比度低和边缘模糊的结直肠息肉图像均有较高的分割精度.

     

    Abstract: Accurate recognition of colorectal images assists doctors in screening for malignant intestinal diseases. Colorectal cancer can be induced by colorectal polyps, ulcerative colitis, and papillary adenoma. Colorectal polyp segmentation can be used to rapidly locate polyps using automated colorectal polyp segmentation technology, saving the time and cost of manual screening and providing patients with valuable treatment time. Therefore, designing an automatic identification and accurate segmentation method is crucial for clinicians to improve diagnosis efficiency. Aiming at the problems in colorectal polyp image segmentation, such as large regional scale variation, variable location, blurred edges, and low contrast between polyps and normal tissues, which lead to low accuracy of lesion segmentation and artifacts on the segmentation boundary, an adaptive network based on the Swin Transformer and graphical reasoning is proposed. First, the Swin Transformer encoder is used to extract the global context information of the input image layer by layer, weaken the interference of background information, and analyze the salient characteristics of the lesion area on a multiscale. Second, a global and local feature interaction module is proposed to enhance the spatial perception ability of the network on complex lesions, highlight the key position information of the target to be segmented, and accurately locate the target. Third, a region-guided graph inference module is used to mine the higher-order dominant relationship between prior information in the way of graph cyclic recurrence to promote the transmission of information between graphs. Fourth, an edge constraint graph inference module oriented to edge details is designed to integrate edge details and improve the segmentation effect and segmentation precision. Experiments were performed on the CVC-ClinicDB, Kvasir, CVC-ColonDB, and ETIS datasets. The Dice coefficients were 0.939, 0.926, 0.810, and 0.788, respectively, and the average intersection ratios were 0.889, 0.879, 0.731, and 0.710, respectively. The mean absolute errors were 0.006, 0.017, 0.030, and 0.012, respectively. Compared with the SSformer method based on the transformer structure, the Dice coefficients are increased by 2.3%, 0.1%, 3.8%, and 2.1%, respectively, and the average crossover ratio is increased by 1.6%, 0.1%, 3.4%, and 1.2%, respectively. The overall segmentation performance of the algorithm test is better than that of the existing method. The simulation results show that the image segmentation accuracy of colorectal polyps with complex shapes and structures, low contrast, and blurred edges is high.

     

/

返回文章
返回