基于改进ADNet网络模型的低剂量CT图像降噪方法

Low-dose CT image noise reduction method based on improved ADNet network model

与正常剂量计算机断层扫描成像(CT)相比,低剂量CT成像可以有效减少X射线对身体的辐射,但因此产生的噪声会显著降低CT成像质量。传统的神经网络由于提取通道单一,影响了图像的特征提取,不利于低剂量CT图像的降噪。分析了基于双注意力机制和记忆与高频特征融合的神经网络图像降噪方法。实验结果表明,与目前常用的3种典型网络相比,该模型避免CT图像过度平滑,可有效保留图像细节纹理。与ADNet网络模型相比在结构相似性上提升了0.0055,峰值信噪比上提升了0.2707。

Compared with normal dose computed tomography(CT)imaging,low-dose CT imaging can effectively reduce the radiation of X-rays to the body,but the resulting noise will significantly reduce the quality of CT imaging and thus affect the doctor′s diagnosis.Because of the single extraction channel,the traditional neural network affects the image feature extraction,which is not conducive to the noise reduction of low-dose CT images.This paper analyzes an image denoising method based on dual attention mechanism and fusion of memory and high frequency features.The experimental results show that,compared with the three typical networks commonly used at present,the model can avoid excessive smoothing of CT images,and can effectively preserve the image texture details.Compared with the ADNet network model,the structure similarity is improved by 0.0055,and the peak signal-to-noise ratio is improved by 0.2707.