摘要
为改善基层医疗机构儿童肺炎诊疗水平,提高基层医生分析临床医学影像的效率和质量,提出了一种基于Vision Transformer(ViT)的小儿肺炎辅助诊断模型。首先利用ResUNet对儿童胸片进行肺区域分割,将左右肺区域从胸片中分割出来以降低其他组织对肺炎诊断的干扰。然后,将分割后的图像输入改进的混合ViT模型进行诊断,该模型使用传统卷积神经网络的特征映射作为Transformer的输入,并在卷积神经网络中引入自注意力机制,增强卷积以加强其获取全局相关性的能力。最后,对卷积神经网络的骨干网络和Transformer模型进行端到端的训练,使模型能够达到良好的图像分类结果。在Chest X-Ray Images肺炎标准数据集上进行了实验,实验结果表明,所提模型的肺炎识别准确率、精确率和召回率分别达到97.27%、97.69%和98.60%。即该模型具有较好的可行性,可使基层儿童肺炎的临床诊断准确率得到很大提升。
To improve the diagnosis and treatment level of pneumonia in children in primary medical institutions and doctors’efficiency and quality in analyzing clinical medical images,an auxiliary diagnosis model of pneumonia in children,based on the Vision Transformer(ViT),is proposed.First,ResUNet is used to segment the lung region in the chest film of children,and the left and right lung regions are separated from the chest film to mitigate the interference of other tissues during pneumonia diagnosis.Further,the segmented image is input into the improved hybrid ViT model for diagnosis.This model uses the feature map of the traditional convolutional neural network(CNN)as the input of the Transformer and introduces the selfattention mechanism into the CNN to improve convolution to enhance its ability to obtain global correlation.Finally,the backbone network of the CNN and Transformer model are trained endtoend so that the proposed model can achieve good image classification results.Experiments were conducted on the Chest XRay Images pneumonia standard dataset.The experimental results show that the accuracy,precision,and recall of the proposed model for pneumonia recognition reach 97.27%,97.69%,and 98.60%respectively.In other words,the model has good feasibility and can significantly improve the clinical diagnosis accuracy of pneumonia in children at the grassroot level.
作者
赵爽
魏国辉
赵文华
马志庆
Zhao Shuang;Wei Guohui;Zhao Wenhua;Ma Zhiqing(Laboratory Management Office,Shandong University of Traditional Chinese Medicine,Jinan 250355,Shandong,China;College of Intelligence and Information Engineering,Shandong University of Traditional Chinese Medicine,Jinan 250355,Shandong,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2023年第4期183-190,共8页
Laser & Optoelectronics Progress
基金
山东省研究生教育质量提升计划课题(SDYJG1943)。
