融合双重注意力网络的儿童骨龄评估方法  

Pediatric bone age assessment method combined with dual attention network

作　　者：张鑫[1] 张俊华[1] 张帅[1] Zhang Xin;Zhang Junhua;Zhang Shuai(School of Information Science&Engineering,Yunnan University,Kunming 650500,China)

机构地区：[1]云南大学信息学院,昆明650500

出　　处：《计算机应用研究》2022年第11期3509-3515,共7页Application Research of Computers

基　　金：国家自然科学基金资助项目(62063034,61841112);云南大学研究生实践创新项目(2021Z50)。

摘　　要：骨龄评估是一种检测儿童内分泌与生长发育异常的常用方法,但深度学习方法中低质量手部X射线图像降低最终评估精度。针对该问题,提出一种增加手部X射线图像感兴趣区域面积的对齐网络,该网络以Swin Transformer结构作为主干网络学习图像手部相似性并取得仿射系数,且在训练过程中无须进行大规模手部标注。在骨龄评估网络中,针对高效通道注意力和空间注意力机制改进,提出双池化高效通道注意力和非对称卷积空间注意力方法,将这两种方法以双重注意力形式与Xception网络相结合提出DA-Xception。在RSNA数据集上进行测试,该骨龄评估方法达到5.37个月的平均绝对误差,相较于其他深度学习方法可更充分提取特征,优化评估结果。Bone age assessment is a common method to detect endocrine and growth abnormalities in children.But in deep learning methods,low-quality hand X-ray images reduce the final evaluation accuracy.To solve this problem,this paper proposed an alignment network to increase the area of interest in hand X-ray images.This network used the Swin Transformer structure as the backbone network to learn image hand similarity and obtain affine coefficients and did not require large-scale hand annotation during the training process.In the bone age assessment network,for the improvement of efficient channel attention and spatial attention mechanism,this paper proposed dual-pool efficient channel attention and asymmetric convolution spatial attention method and combined these two methods in the form of dual attention and Xception network to propose DA-Xception.When tested on the RSNA dataset,this bone age assessment method achieved a mean absolute error of 5.37 months.Compared with other deep learning methods,this method can fully extract features and optimize the evaluation results.

关 键 词：骨龄评估 X射线图像对齐 双重注意力 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]