C-Hybrid-NET: A self-attention-based COVID-19 screening model based on concatenated hybrid 2D-3D CNN features from chest X-ray images - Université de Bourgogne Accéder directement au contenu
Article Dans Une Revue Multimedia Tools and Applications Année : 2024

C-Hybrid-NET: A self-attention-based COVID-19 screening model based on concatenated hybrid 2D-3D CNN features from chest X-ray images

Résumé

The outbreak of novel coronavirus (2019-nCOV, commonly known as COVID-19) was declared a global pandemic by the World Health Organization (WHO) in March 2020. An accurate diagnostic tool is needed to test for COVID-19, as this disease is highly contagious and can be fatal in some patients. The use of radiological methods to detect the early stages of COVID-19 pneumonia remains one of the greatest challenges facing the global medical community. With the development of deep learning methods, COVID-19 screening can be performed using Convolutional Neural Networks (CNNs) and self-attention mechanisms from chest X-rays. Recently few attempts have been made to use 3D CNNs to learn spatial contextual information from 2D X-ray images due to their computational complexity. Inspired by the spectacular performance of self-attention mechanisms and 2D/3D CNNs, we develop a self-attention-based hybrid 2D/3D CNN model, called C-Hybrid-NET, to extract and learn complex and more contextual feature representations while maintaining diagnostic performance. In particular, we apply transfer learning techniques to improve the generalization ability of the model in the presence of unbalanced data distribution. In addition, we perform feature concatenation on high-level activations of multiple pre-trained 2D CNNs to improve the discriminative power of the features. We also incorporate the self-attention mechanism into our C-Hybrid-NET model to capture semantic associations between multi-level features and improve its recognition ability. Experimental results on a collected chest X-ray dataset show that our diagnostic framework performs well despite the lack of training samples, achieving an overall accuracy of 99.76%, a sensitivity of 96.82%, and a specificity of 99.94% for binary classification, and 97.91%, 99.55%, and 99.00% for multi-class classification, respectively.
Fichier non déposé

Dates et versions

hal-04647377 , version 1 (14-07-2024)

Identifiants

Citer

Khaled Bayoudh, Fayçal Hamdaoui, Abdellatif Mtibaa. C-Hybrid-NET: A self-attention-based COVID-19 screening model based on concatenated hybrid 2D-3D CNN features from chest X-ray images. Multimedia Tools and Applications, 2024, ⟨10.1007/s11042-024-19800-w⟩. ⟨hal-04647377⟩
0 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More