带有干扰挖掘的息肉分割算法
Polyp segmentation algorithm with interference mining
张威威 1刘军清 1康维1
作者信息
摘要
息肉作为身体表面生长的异常组织,存在病变的可能,需要长期观察监测.而息肉分割是观察监测以及后续治疗的重要手段.现有研究表明,精确的息肉分割主要难度在于息肉边界对比度低、外观变化多以及多息肉存在.针对此,在聚焦多尺度问题的基础上,受生物视觉的影响提出并实现了一个带有干扰挖掘的息肉分割算法,该网络模拟生物视觉发现目标的过程,以全局角度从带有干扰的图像中定位潜在目标,然后进入识别过程,以通过聚焦模糊区域来挖掘目标信息得到初步预测图,将初步预测的结果作为注意力图引导浅层特征,最后,将细化后的特征图输入至多尺度残差推理模块,产生最终预测.
Abstract
Polyps are abnormal tissue growths on the surface of the body that have the potential to become diseased and require long-term observation and monitoring.Polyp segmentation is an important tool for monitoring and subsequent treatment.Existing studies have shown that the main difficulties in accurate polyp segmentation are the low contrast of polyp boundaries,the many variations in appearance,and the presence of multiple polyps.To address this,a polyp seg-mentation algorithm with interference mining is proposed and implemented on the basis of the focusing multi-scale problem,influenced by biological vision,which simulates the process of tar-get discovery by biological vision,locates potential targets from images with interference from a global perspective,and then enters into the recognition process in order to mine the target infor-mation to get a preliminary prediction map by focusing on the fuzzy region,and then the pre-liminary prediction results as an attention map to guide the shallow features,and finally,the re-fined feature map is fed into the multiscale residual inference module to produce the final pre-diction.
关键词
息肉分割/干扰挖掘/定位聚焦/多尺度残差推理Key words
Polyp segmentation/interference mining/positioning focusing/multiscale residual inference引用本文复制引用
出版年
2024
NETL
NSTL
万方数据