基于深度学习的慢性阻塞性肺病的诊断模型研究

慢性阻塞性肺病(COPD)是一种常见的以持续气流受限为特征的慢性呼吸道疾病,具有很高的发病率和死亡率。目前临床上对COPD的诊断方式十分复杂,不仅耗时且有创或有辐射伤害,不适用于日常筛查。本研究设计了一种基于深度学习的COPD诊断模型。首先,将RespiratoryDatabase @ TR多媒体呼吸数据库中42位COPD患者的肺音数据和来自天津大学胸科医院的24位COPD患者以及37位健康受试者的临床采集肺音数据相结合,分别运用高通滤波器和基于集合经验模态分解(EEMD)及小波熵的去噪算法进行去噪处理,然后通过归一化、交叠剪切、数据扩增完成预处理过程;然后利用二阶谱分析技术提取肺音特征;最后,将特征输入到改进的19层卷积神经网络模型中,实现健康受试者与COPD患者的二分类。实验结果表明,所提出的模型能够有效诊断COPD,其准确度、敏感度、特异性、F1分数和Kappa系数分别达到了98.93%、98.47%、99.41%、98.95%和97.86%,且由于采用了双中心数据并进行了去噪处理,模型可靠性更高,具有重要的临床意义。

Diagnosis Model of Chronic Obstructive Pulmonary Disease Based on Deep Learning

Chronic obstructive pulmonary disease (COPD) is a common chronic respiratory disease characterized by continuous airflow restriction, with high morbidity and mortality. At present, clinical diagnosis methods of COPD are very complex, not only time-consuming and invasive or radioactive, and not suitable for daily screening. Therefore, a COPD diagnosis model based on deep learning was designed in this study. Firstly, the lung sound from 42 COPD patients from RespiratoryDatabase@TR multimedia respiratory database were combined with the clinically collected lung sound from 24 COPD patients and 37 healthy subjects from Chest Hospital of Tianjin University, high-pass filter and denoising algorithm based on ensemble empirical mode decomposition (EEMD) and wavelet entropy was used for denoising. Secondly, the pre-processing process was completed through normalization, overlapping shear and data amplification. Thirdly, bispectrum analysis was used to extract the lung sound features. Finally, these features were input into an improved 19-layer convolutional neural network model to achieve the binary classification of healthy subjects and COPD patients. Experimental results showed that the proposed model could effectively diagnose COPD. The accuracy, sensitivity, specificity, F1 score, and Kappa score reached 98.93%, 98.47%, 99.41%, 98.95%, and 97.86%, respectively. Moreover, due to the use of bicentric data and denoising process, the model has higher reliability and is of important clinical significance.