| 摘 要: 医学图像处理对医疗诊断具有重要价值,在儿童牙科疾病的诊断中,通常需要对全景 X光片进行分割来准确分析牙齿疾病。针对儿童牙齿全景 X线摄影分割中存在的牙齿解剖结构复杂导致传统方法精度不足的问题,提出一种基于 Transformer的网络算法。该算法采用双路径注意力机制,通过在编码器阶段交替执行空间自注意力与通道自注意力聚合,有效捕捉儿童牙齿的形态特征。解码器设计自适应特征融合模块,实现深层语义信息与浅层细节特征的自适应融合。实验结果表明,该算法在儿童牙齿全景 X线的分割任务中取得更好的结果。 |
| 关键词: 深度学习 牙齿全景 X线 医学图像分割 Transformer |
|
中图分类号: TP391
文献标识码: A
|
|
| Children’s Dental Panoramic Radiographs Based on Transformer |
|
CHEN Lingxiao, CHEN Xiaodiao
|
(Department of Computer Science, Hangzhou Dianzi University, Hangzhou 310018, China)
ingxiao@hdu.edu.cn; xiaodiao@hdu.edu.cn
|
| Abstract: Medical image processing holds significant value in clinical diagnosis, particularly in pediatric dentistry where accurate analysis often requires the segmentation of panoramic radiographs. To address the limitations of traditional methods in segmenting pediatric dental panoramic radiographs, which stem from the complex anatomical structures of children’s teeth, this study proposes a Transforme-r based network algorithm. The model incorporates a dua-l path attention mechanism that alternately performs spatial sel-f attention and channel sel-f attention aggregation during the encoder phase, effectively capturing the morphological features of children’s teeth. Additionally, the decoder is designed with an adaptive feature fusion module to achieve adaptive integration of deep semantic information and shallow detail features. The experimental results demonstrate that the proposed method achieves superior performance in the segmentation of pediatric dental panoramic radiographs. |
| Keywords: deep learning dental panoramic radiographs medical image segmentation Transformer |
