Scholarlink

<p>The Coronavirus (COVID-19) crisis, starting from the novel Covid SARS-CoV-2, significantly affected the world since its beginning in late 2019. It has caused a huge death toll, upset economies, stressed medical services frameworks, and changed the way people live and interact with one another. The spread of COVID-19 was rapid and global, and its impact has been devastating in terms of human lives lost. Deep learning (DL) helped researchers to find speedy solutions to overcome this crisis. Different DL algorithms have been used to develop computer vision models capable of detecting COVID-19 from medical imaging datasets, such as chest X-rays (XR) and Computed tomography (CT). In our paper, we proposed a COVID-19 diagnosis system combining two cascaded networks for classification and segmentation of CT images. The classification process is based on using parallel paths to extract features from different level. Four models: FPRCNN-1, FPRCNN-2, FPRCNN-3, and FPRCNN-4 has been proposed in the classification stage. All these CNN models are based on deep learning and consist of five parallel CNN layers. We applied different sizes of M-Blocks which consists of some layers and skipped connection. In the segmentation process, we proposed a modified U-Net (MU-Net) architecture using different CT slices from 20 patients. We tested the performance of the classification models, and we found that FPRCNN-4 could achieve the highest performance results among other models and literature with an accuracy value of 98.8% after applying full residual on it. The segmentation performance achieved a training dice score of 92.67% and testing dice score of 82.18%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Covid-19 lung infection diagnosis model using parallel and cascaded CNN architecture based on M-blocks and modified U-Net

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Mahmoud Khaled Abd-Ellah,
  • Asmaa Mohamed,
  • Gerges M. Salama

摘要

The Coronavirus (COVID-19) crisis, starting from the novel Covid SARS-CoV-2, significantly affected the world since its beginning in late 2019. It has caused a huge death toll, upset economies, stressed medical services frameworks, and changed the way people live and interact with one another. The spread of COVID-19 was rapid and global, and its impact has been devastating in terms of human lives lost. Deep learning (DL) helped researchers to find speedy solutions to overcome this crisis. Different DL algorithms have been used to develop computer vision models capable of detecting COVID-19 from medical imaging datasets, such as chest X-rays (XR) and Computed tomography (CT). In our paper, we proposed a COVID-19 diagnosis system combining two cascaded networks for classification and segmentation of CT images. The classification process is based on using parallel paths to extract features from different level. Four models: FPRCNN-1, FPRCNN-2, FPRCNN-3, and FPRCNN-4 has been proposed in the classification stage. All these CNN models are based on deep learning and consist of five parallel CNN layers. We applied different sizes of M-Blocks which consists of some layers and skipped connection. In the segmentation process, we proposed a modified U-Net (MU-Net) architecture using different CT slices from 20 patients. We tested the performance of the classification models, and we found that FPRCNN-4 could achieve the highest performance results among other models and literature with an accuracy value of 98.8% after applying full residual on it. The segmentation performance achieved a training dice score of 92.67% and testing dice score of 82.18%.