tetano
Editor, Senior Moderator
Comput Biol Med
. 2022 Jul 19;149:105806.
doi: 10.1016/j.compbiomed.2022.105806. Online ahead of print.
A dual-stage deep convolutional neural network for automatic diagnosis of COVID-19 and pneumonia from chest CT images
Farhan Sadik[SUP] 1 [/SUP], Ankan Ghosh Dastider[SUP] 1 [/SUP], Mohseu Rashid Subah[SUP] 1 [/SUP], Tanvir Mahmud[SUP] 1 [/SUP], Shaikh Anowarul Fattah[SUP] 2 [/SUP]
Affiliations
Abstract
In the Coronavirus disease-2019 (COVID-19) pandemic, for fast and accurate diagnosis of a large number of patients, besides traditional methods, automated diagnostic tools are now extremely required. In this paper, a deep convolutional neural network (CNN) based scheme is proposed for automated accurate diagnosis of COVID-19 from lung computed tomography (CT) scan images. First, for the automated segmentation of lung regions in a chest CT scan, a modified CNN architecture, namely SKICU-Net is proposed by incorporating additional skip interconnections in the U-Net model that overcome the loss of information in dimension scaling. Next, an agglomerative hierarchical clustering is deployed to eliminate the CT slices without significant information. Finally, for effective feature extraction and diagnosis of COVID-19 and pneumonia from the segmented lung slices, a modified DenseNet architecture, namely P-DenseCOVNet is designed where parallel convolutional paths are introduced on top of the conventional DenseNet model for getting better performance through overcoming the loss of positional arguments. Outstanding performances have been achieved with an F[SUB]1[/SUB] score of 0.97 in the segmentation task along with an accuracy of 87.5% in diagnosing COVID-19, common pneumonia, and normal cases. Significant experimental results and comparison with other studies show that the proposed scheme provides very satisfactory performances and can serve as an effective diagnostic tool in the current pandemic.
Keywords: COVID-19; Classification; Clustering; Convolutional neural network; Deep learning; Segmentation.
. 2022 Jul 19;149:105806.
doi: 10.1016/j.compbiomed.2022.105806. Online ahead of print.
A dual-stage deep convolutional neural network for automatic diagnosis of COVID-19 and pneumonia from chest CT images
Farhan Sadik[SUP] 1 [/SUP], Ankan Ghosh Dastider[SUP] 1 [/SUP], Mohseu Rashid Subah[SUP] 1 [/SUP], Tanvir Mahmud[SUP] 1 [/SUP], Shaikh Anowarul Fattah[SUP] 2 [/SUP]
Affiliations
- PMID: 35994932
- DOI: 10.1016/j.compbiomed.2022.105806
Abstract
In the Coronavirus disease-2019 (COVID-19) pandemic, for fast and accurate diagnosis of a large number of patients, besides traditional methods, automated diagnostic tools are now extremely required. In this paper, a deep convolutional neural network (CNN) based scheme is proposed for automated accurate diagnosis of COVID-19 from lung computed tomography (CT) scan images. First, for the automated segmentation of lung regions in a chest CT scan, a modified CNN architecture, namely SKICU-Net is proposed by incorporating additional skip interconnections in the U-Net model that overcome the loss of information in dimension scaling. Next, an agglomerative hierarchical clustering is deployed to eliminate the CT slices without significant information. Finally, for effective feature extraction and diagnosis of COVID-19 and pneumonia from the segmented lung slices, a modified DenseNet architecture, namely P-DenseCOVNet is designed where parallel convolutional paths are introduced on top of the conventional DenseNet model for getting better performance through overcoming the loss of positional arguments. Outstanding performances have been achieved with an F[SUB]1[/SUB] score of 0.97 in the segmentation task along with an accuracy of 87.5% in diagnosing COVID-19, common pneumonia, and normal cases. Significant experimental results and comparison with other studies show that the proposed scheme provides very satisfactory performances and can serve as an effective diagnostic tool in the current pandemic.
Keywords: COVID-19; Classification; Clustering; Convolutional neural network; Deep learning; Segmentation.