A deep learning architecture for multi-class lung diseases classification using chest X-ray (CXR) images
A deep learning architecture for multi-class lung diseases classification using chest X-ray (CXR) images
复制标题
DOI：
10.1016/j.aej.2022.10.053
复制
发表时间：
2022-11-02
影响因子：
6.8
通讯作者：
中科院分区：
文献类型：
--
作者：
2019年，全球经历了新冠肺炎疫情的快速爆发，在全球范围内造成了令人担忧的局面。该病毒以呼吸系统为目标，引起肺炎，并伴有其他症状，如疲劳、干咳和发烧，这些症状可能被误诊为肺炎、肺癌或结核病。因此，COVID-19的早期诊断至关重要，因为该疾病可能导致患者死亡。胸部X射线（CXR）通常用于医疗保健部门，可以提供快速和精确的诊断。深度学习算法在肺部疾病检测和分类方面具有非凡的能力。他们促进和加快诊断过程，并为医生节省时间。本文提出了一种深度学习（DL）架构，用于肺炎、肺癌、结核病（TB）、肺混浊和最近的COVID-19的多类分类。对3615张COVID-19、6012张肺混浊、5870张肺炎、20，000张肺癌、1400张结核病和10，192张正常图像的大量CXR图像进行了大小调整、标准化和随机分割，以符合DL要求。在分类方面，我们使用了一个预训练的模型VGG 19，然后是三个卷积神经网络（CNN）块作为分类阶段的特征提取和全连接网络。实验结果表明，我们提出的VGG 19 + CNN优于其他现有的工作，准确率为96.48%，召回率为93.75%，精确率为97.56%，F1得分为95.62%，曲线下面积（AUC）为99.82%。所提出的模型提供了上级性能，使医疗从业者能够更快速有效地诊断和治疗患者。
In 2019, the world experienced the rapid outbreak of the Covid-19 pandemic creating an alarming situation worldwide. The virus targets the respiratory system causing pneumonia with other symptoms such as fatigue, dry cough, and fever which can be mistakenly diagnosed as pneumonia, lung cancer, or TB. Thus, the early diagnosis of COVID-19 is critical since the disease can provoke patients’ mortality. Chest X-ray (CXR) is commonly employed in healthcare sector where both quick and precise diagnosis can be supplied. Deep learning algorithms have proved extraordinary capabilities in terms of lung diseases detection and classification. They facilitate and expedite the diagnosis process and save time for the medical practitioners. In this paper, a deep learning (DL) architecture for multi-class classification of Pneumonia, Lung Cancer, tuberculosis (TB), Lung Opacity, and most recently COVID-19 is proposed. Tremendous CXR images of 3615 COVID-19, 6012 Lung opacity, 5870 Pneumonia, 20,000 lung cancer, 1400 tuberculosis, and 10,192 normal images were resized, normalized, and randomly split to fit the DL requirements. In terms of classification, we utilized a pre-trained model, VGG19 followed by three blocks of convolutional neural network (CNN) as a feature extraction and fully connected network at the classification stage. The experimental results revealed that our proposed VGG19 + CNN outperformed other existing work with 96.48 % accuracy, 93.75 % recall, 97.56 % precision, 95.62 % F1 score, and 99.82 % area under the curve (AUC). The proposed model delivered superior performance allowing healthcare practitioners to diagnose and treat patients more quickly and efficiently.
登录
查看更多内容
影响因子：
3.6
作者：
Hooda, Rahul;Mittal, Ajay;Sofat, Sanjeev
通讯作者：
Sofat, Sanjeev
影响因子：
8.5
作者：
Chandra, Tej Bahadur;Verma, Kesari;Netam, Satyabhuwan Singh
通讯作者：
Netam, Satyabhuwan Singh
影响因子：
10.6
作者：
Jaeger, Stefan;Karargyris, Alexandros;McDonald, Clement J.
通讯作者：
McDonald, Clement J.
影响因子：
10.6
作者：
Candemir, Sema;Jaeger, Stefan;McDonald, Clement J.
通讯作者：
McDonald, Clement J.
影响因子：
8.7
作者：
ALzubi, Jafar A.;Bharathikannan, Balasubramaniyan;Thaventhiran, Chandrasekar
通讯作者：
Thaventhiran, Chandrasekar