Udayaraju Pamula, Narayana T Venkata, Vemparala Sri Harsha, Srinivasarao Chopparapu, Raju BhV S R K
Department of CSE, SRKR Engineering College, Affiliated to JNTUK, Bhimavaram, AP, India.
Department of ECE, SRKR Engineering College, Affiliated to JNTUK, Bhimavaram, AP, India.
Measur Sens. 2023 Jun;27:100735. doi: 10.1016/j.measen.2023.100735. Epub 2023 Mar 20.
COVID-19 is one of the dangerous viruses that cause death if the patient doesn't identify it in the early stages. Firstly, this virus is identified in China, Wuhan city. This virus spreads very fast compared with other viruses. Many tests are there for detecting this virus, and also side effects may find while testing this disease. Corona-virus tests are now rare; there are restricted COVID-19 testing units and they can't be made quickly enough, causing alarm. Thus, we want to depend on other determination measures. There are three distinct sorts of COVID-19 testing systems: RTPCR, CT, and CXR. There are certain limitations to RTPCR, which is the most time-consuming technique, and CT-scan results in exposure to radiation which may cause further diseases. So, to overcome these limitations, the CXR technique emits comparatively less radiation, and the patient need not be close to the medical staff. COVID-19 detection from CXR images has been tested using a diversity of pre-trained deep-learning algorithms, with the best methods being fine-tuned to maximize detection accuracy. In this work, the model called GW-CNNDC is presented. The Lung Radiography pictures are portioned utilizing the Enhanced CNN model, deployed with RESNET-50 Architecture with an image size of 255*255 pixels. Afterward, the Gradient Weighted model is applied, which shows the specific separation regardless of whether the individual is impacted by Covid-19 affected area. This framework can perform twofold class assignments with exactness and accuracy, precision, recall, F1-score, and Loss value, and the model turns out proficiently for huge datasets with less measure of time.
新型冠状病毒肺炎（COVID-19）是一种危险的病毒，如果患者在早期未识别出来，可能会导致死亡。首先，这种病毒在中国武汉市被发现。与其他病毒相比，这种病毒传播速度非常快。有许多检测这种病毒的方法，并且在检测这种疾病时也可能会发现副作用。冠状病毒检测现在很罕见；新冠病毒检测单位有限，而且无法快速生产出来，这引起了恐慌。因此，我们想依靠其他判定措施。有三种不同类型的新冠病毒检测系统：逆转录聚合酶链反应（RTPCR）、计算机断层扫描（CT）和胸部X线摄影（CXR）。RTPCR有一定局限性，它是最耗时的技术，而CT扫描会导致辐射暴露，这可能会引发其他疾病。所以，为了克服这些局限性，CXR技术辐射相对较少，并且患者无需靠近医护人员。已经使用多种预训练的深度学习算法对从CXR图像中检测新冠病毒进行了测试，最好的方法经过了微调以最大化检测准确率。在这项工作中，提出了名为GW-CNNDC的模型。利用增强型卷积神经网络（CNN）模型对肺部X光照片进行分割，该模型采用RESNET-50架构，图像大小为255×255像素。然后，应用梯度加权模型，无论个体是否受到新冠病毒感染区域的影响，该模型都能显示出特定的区分。这个框架能够以精确性、准确性、精确率、召回率、F1分数和损失值进行二元分类任务，并且该模型在处理大型数据集时能够高效运行，用时较少。
