CODE-NET++: An Attention-Guided Deep Learning Framework with Grad-CAM-based Explainability for COVID-19 Detection Using Chest X-ray Images

Authors

• Fareesa Amina, Research scholar, Department of Electronics and Communication Engineering, Bharatiya Engineering Science & Technology Innovation University (BESTIU), Gownivaripalli, Gorantla Mandal, Sri Satya Sai District, Andhra Pradesh, IndiaFollow
• Dr Krishnanaik Vankdoth, Professor, Department of Electronics and Communication Engineering, Chaitanya Deemed to be University, Hyderabad, IndiaFollow

Corresponding Author

Dr. Krishnanaik Vankdoth Professor(ECE), Chaitanya Deemed to be University, Hyderabad, India Email: krishnanaik.ece@gmail.com

Subject Area

Electronics and Communication Engineering

Article Type

Special Issue Original Study

Abstract

Chest radiograph imaging has emerged as a practical and scalable diagnostic modality for respiratory diseases, including COVID-19. However, accurate discrimination of COVID-19 manifestations from other pulmonary abnormalities remains challenging because of low contrast, imaging noise, and overlapping radiographic patterns. This work presents CODE-NET++, an enhanced attention-guided deep learning framework with Grad-CAM-based explainability for reliable COVID-19 detection using chest X-ray images. The proposed framework integrates adaptive trilateral filtering for image enhancement, Reverse Edge Attention Network (RE-Net) for lesion-aware segmentation, and an Enhanced LinkNet architecture with dilated convolutions for multiscale feature extraction and classification. Grad-CAM-based explainable artificial intelligence visualization is incorporated to provide clinically interpretable heat maps highlighting class-discriminative lung regions. Experimental evaluation conducted on the COVIDx dataset demonstrates that the proposed CODE-NET++ framework achieves a classification accuracy of 99.75%, outperforming several state-of-the-art deep learning models. The framework also achieved superior segmentation performance with Dice Index and Jaccard Index values of 93.56% and 96.79%, respectively. The proposed CODE-NET++ framework effectively combines preprocessing, lesion-aware segmentation, multiscale classification, and explainable visualization into a unified architecture. The obtained results confirm strong diagnostic accuracy, robustness, and interpretability, making the framework suitable for reliable clinical deployment.

Keywords

COVID-19 detection, Chest X-ray, Deep learning, Attention mechanism, Medical image segmentation, Explainable AI

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 License.

Recommended Citation

Amina, Fareesa and Vankdoth, Dr Krishnanaik (2026) "CODE-NET++: An Attention-Guided Deep Learning Framework with Grad-CAM-based Explainability for COVID-19 Detection Using Chest X-ray Images," Mansoura Engineering Journal: Vol. 51 : Iss. 4 , Article 14.
Available at: https://doi.org/10.58491/2735-4202.3441