Publication: CCVNet: COVID-19 Detection in Chest X-ray Imaging based on Convolutional Neural Network
0
0
Issued Date
2023-05-12
Resource Type
Scopus ID
2-s2.0-85178046279
Journal Title
ACM International Conference Proceeding Series
Start Page
1
End Page
5
Rights Holder(s)
SCOPUS
Bibliographic Citation
ACM International Conference Proceeding Series (2023) , 1-5
Suggested Citation
Choomueang W., Withoonchatri C., Janwong P., Sa-Ing V. CCVNet: COVID-19 Detection in Chest X-ray Imaging based on Convolutional Neural Network. ACM International Conference Proceeding Series (2023) , 1-5. 5. doi:10.1145/3608298.3608299 Retrieved from: https://hdl.handle.net/20.500.14740/20216
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
COVID-19 is a violent disease that rapidly transmits from human to human until the world pandemic. The preliminary screening for this disease is the reverse transcription-polymerase chain reaction (RT-PCR) testing, but the chest X-ray imaging in the hospital was more accurate. However, the doctors or specialists must visualize the organs on images and search for the indication of symptoms in many cases per day. This is the main problem that can be the tiredness of making the wrong diagnosis. Therefore, this research will propose an improved convolutional neural (CNN) network architecture for assisting the detection between COVID-19 and normal patients in improving chest X-ray images that is called CCVNet. Our research gathers the three public datasets to create a balanced chest X-ray dataset for using training, validation, and testing CNN models. The balanced dataset, we improve by removing the unusual makers to increase the detection accuracy, and then augment by using three simulated criterions to create 30,576 images for each COVID-19 and normal cases. The first experiment compared the cutting-edge CNN models to use the analysis of medical imaging that consists of VGG16, RestNet50, and Xception. From the first results, VGG16 represented 93.1% accuracy and was the best model with the least overfitting results. The second experiment improved the VGG16 architecture in many layers and setting parameters to represent the 94.4% accuracy of the training process. In addition, the proposed model was evaluated by using the confusion matrix to measure the accuracy, sensitivity, specificity, precision, F1 score, and receiver operating characteristic curve. From these evaluation results, the improved model shows an average accuracy of corrected detection of about 94.4%. Therefore, the proposed CCVNet model can assist in diagnosing the COVID-19 and normal patients in the chest X-ray images that remove the unusual markers.