DDCNNC: Dilated and depthwise separable convolutional neural Network for diagnosis COVID-19 via chest X-ray images

Abstract

Purpose As of December 21, 2020, a total of 77,670,400 cases of coronavirus disease 2019 (COVID-19) have been confirmed worldwide, 53,825,243 cases have been cured and 1,693,253 cases have died. Among the diagnostic methods of COVID-19, chest X-ray images have the advantages of fast imaging, low cost and high accuracy of single plane lesions recognition. The current COVID-19 detection models have shortcomings such as weak robustness, unreliable generalization ability, and long training time. Methods To solve the above problems, our team proposed two novel frameworks and five methods to diagnose COVID-19 based on chest X-ray images. (i) A novel framework – depthwise separable convolutional neural network (DCNN), and we tested Three methods, viz., using LeNet-5, VGG-16, and ResNet-18 as backbones. (ii) A novel framework – dilated and depthwise separable convolutional neural network (DDCNN), and we tested Two methods, viz., using VGG-16 and ResNet-18 as backbones. Results Experiment results show that our models not only improve the detection accuracy, but also reduce the training time. Conclusions Our methods are superior to state-of-the-art methods in both above aspects.

Artificial Intelligence for Forecasting the Prevalence of COVID-19 Pandemic: An Overview

“…They succeeded in developing a single architecture of the model that can be used to achieve both targets. Li et al [ 86 ] utilized three other CNN models, namely VGG-16, LeNet-5 and ResNet-18, to detect COVID-19 infection using X-ray images. EfficientNet CNN was used to classify the diagnosed cases into normal, pneumonia andCOVID-19 cases based on their X-ray images [ 87 ].…”

A transfer learning-based deep learning approach for automated COVID-19 diagnosis with audio data

“…Recent ML-based studies related to COVID-19 diagnosis mostly use X-ray or CT scan ( [49], [50], [51], [52]). Although they provide high classification accuracies, they oblige people to go to hospitals for screening.…”

A Hybrid Pipeline for Covid-19 Screening Incorporating Lungs Segmentation and Wavelet Based Preprocessing of Chest X-Rays

“…The CONV RES blocks require the training of fewer than 20,000 parameters. It is worth noting that another type of residual network using depthwise separable atrous convolutions with multi-scale kernels in the residual path, named ‘Dilated and Depthwise separable Convolutional Neural Network (DDCNN)’ has been proposed recently by Li et al [10] for the diagnosis of Covid-19 from CXRs. However, while DDCNN progressively reduces the spatial dimensions of the feature maps as they pass through several pooling layers, the feature maps of CXR-Net Module 2 retain the same dimensions (75 × 85 × 51) of the input image (as downsized by the WST block), with the exception of the feature map produced by the last block, which is of dimensions 75 × 85 × 2, as it is used for binary classification (Fig.…”

“…In fact, as impressive as the results of many studies have been in terms of classification statistics, the resulting saliency maps (usually calculated as Gradient Weighted Class Activation Maps (Grad-CAM) [9]) have been rather disappointing with activation areas often extending over both lung and non-lung regions of the CXRs (see, for example, Figure 10 in [3], Figs. 8-9 in [10], Fig. 5 in [11]), casting doubts on whether the reported classification performance is based on the recognition of particular SARS-CoV-2 features of the lungs texture or some other information (age, body shape, bone structure, sex, race, patient positioning, radiographic projection, etc.)…”