COVLIAS 3.5: integration of attention-based segmentation technique with fuzzy dilated convolutional neural networks for improved classification of chest X-ray scans for multiclass pneumonia diagnosis.
Authors
Affiliations (10)
Affiliations (10)
- Department of Information Technology, Bharati Vidyapeeth's College of Engineering, New Delhi, 110063, India.
- INTI International University, Persiaran Perdana BBN Putra Nilai, 71800, Nilai, Negeri Sembilan, Malaysia.
- Department of ECE, Manav Rachna University, Sector-43, Aravali Hills, Surajkund Road, Faridabad, Haryana, 121001, India.
- Department of Computer Science and Engineering, Technocrats Institute of Technology, Bhopal, 462022, India.
- Department of Computer Science and Engineering, Manipal University Jaipur, Jaipur, India. [email protected].
- Department of Radiology, Azienda Ospedaliero Universitaria (A.O.U.) di Cagliari, Cagliari, Italy.
- Stroke Diagnostic and Monitoring Division, AtheroPoint, Roseville, CA, USA.
- Department of Electrical and Computer Engineering, Idaho State University, Pocatello, ID, USA.
- Global Biomedical Technologies, Inc., Roseville, CA, USA.
- Symbiosis Institute of Technology, Nagpur Campus, Symbiosis International (Deemed University), Pune, India.
Abstract
This study aims to improve pneumonia diagnosis by integrating attention-based U-Net models for lung segmentation with fuzzy logic-enhanced CNNs for classification. This approach addresses the limitations of inadequate modelling of complex spatial relationships in medical images. The underutilization of fuzzy logic with dilated convolutions has restricted the extraction of multiscale features. We utilized 18,608 chest X-ray (CXR) images. Subsequently, these images were segmented using four models namely: U-Net, Attention U-Net, Pruned U-Net and U-Net++. Fuzzy logic system was used to process the segmented data. Additionally, we show that the dilated CNN architecture for improved classification performance. In lung segmentation, our experimental results indicate that Attention U-Net (AU) achieved 1% better mean accuracy, 2% better mean Jaccard and Dice than U-Net, pruned U-Net and U-Net++. In the classification, the model has demonstrated 10% better mean accuracy over augmented U-Net and Attention U-Net based segmented data. ROCs have shown that augmented effect has 15% better AUC in bacterial pneumonia class. Additionally, we saw a 4% improvement with fuzzy logic. The integration of fuzzy dilated CNN with Attention U-Net segmentation presents a 1% better accuracy compared to U-Net. Best AUC achieved was 0.98 in bacterial pneumonia. Our findings underscore the critical role of attention mechanisms and augmentation in enhancing medical image analysis. The integration of Attention U-Net for segmentation and fuzzy dilated CNN for multi-class classification significantly improves diagnostic accuracy and reliability. This approach has the potential to revolutionize pneumonia diagnosis, leading to better patient outcomes and more efficient healthcare delivery.