双信道注意力机制神经网络的肺炎图像分类研究

Research on Pneumonia Image Classification Based on Dual Channel Attention Mechanism Neural Network

摘要: 为提高诊断肺炎患者的时效性和精确性，本文提出了三种基于双信道注意力机制的肺部X光图像检测模型。采用了21,158张肺部X光图像，并对原始图像进行图像增强、几何变化等处理。将预处理后的X光图像分别输入三种模型并检测，最后采用识别的准确度、精确度、灵敏性、F1-Score等指标评估了各个模型的性能。实验结果表明，基于双通道注意力机制的DenseNet201模型效果最优，其准确度、精确度、灵敏度、F1-Score分别为95.112%、96.2%、95.4%、95.9%，其中准确度已超过近年来他人4分类肺炎模型，基于双通道注意力机制的DenseNet201模型有利于提高肺炎图像的分类检测效率，可以应用于快速医学筛查。

Abstract: In order to improve the timeliness and accuracy of the diagnosis of pneumonia patients, this paper proposes three lung X-ray image detection models based on dual channel attention. This paper uses 21,158 lung X-ray images, then these original images are processed with image enhancement and geometric changes. Next, the preprocessed X-ray images are input into three different models for detection. Finally, the performance of each model is evaluated by using evaluation factors such as accuracy, precision, sensitivity, and F1-Score. The results indicate that the DenseNet201 model based on dual channel attention has gained greater performance increasing up on average of 95.112%, 96.2%, 95.4%, and 95.9% for accuracy, precision, sensitivity, and F1-Score, respectively. In terms of accuracy, it has surpassed that of others’ 4 classification pneumonia models in recent years. The DenseNet201 model based on dual channel attention mechanism is beneficial to improve the classification and detection efficiency of pneumonia images, this model can be applied to medical rapid screening.