Scholarlink

<p>Amid the persistent threat of respiratory infections like COVID-19 and community-acquired pneumonia (CAP), the need for cutting-edge diagnostic solutions has never been more urgent. Traditional methods often struggle with overlapping radiographic features and subjective interpretation, leading to diagnostic challenges. Artificial intelligence-powered computer-aided diagnostic (AI-CAD) systems offer a transformative approach by enhancing detection and classification accuracy. In this study, we introduce an innovative Attention-Aided Convolutional Neural Network (CNN) Autoencoder Network (AACNet) to improve automated diagnosis. AACNet employs a two-step methodology: first, a deep convolutional autoencoder with a soft attention gate learns and refines radiographic features in an unsupervised manner. Then, a CNN—built upon architectures like Xception, DenseNet-121, and ResNet-50—utilizes these enhanced feature maps for final classification. AACNet significantly surpasses traditional CNN models, addressing performance plateaus observed in conventional feature extraction. For instance, while DenseNet-121 alone achieves 91.56% accuracy, integrating it with AACNet boosts accuracy to 98.57%. Similarly, ResNet-50 and Xception paired with AACNet reach 97.74% and 98.25%, respectively. By mitigating overfitting and improving classification robustness, AACNet enhances AI-driven diagnostic precision. This advancement has profound implications for clinical practice, facilitating earlier detection and intervention, which are crucial for reducing mortality and healthcare burdens. Beyond COVID-19 and CAP, AACNet’s adaptability extends to diagnosing complex conditions such as cancer and brain tumors, underscoring its potential as a breakthrough in AI-powered medical imaging.</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

AACNet: attention aided CNN-autoencoder network for precise categorization of respiratory conditions from HRCT scans

  • 期刊论文

  • 发表日期：

    2025年8月20日

查看全文
  • Anchita Kakati,
  • Upasana Bhattacharjya,
  • Jyoti Prakash Medhi,
  • Kandarpa Kumar Sarma

摘要

Amid the persistent threat of respiratory infections like COVID-19 and community-acquired pneumonia (CAP), the need for cutting-edge diagnostic solutions has never been more urgent. Traditional methods often struggle with overlapping radiographic features and subjective interpretation, leading to diagnostic challenges. Artificial intelligence-powered computer-aided diagnostic (AI-CAD) systems offer a transformative approach by enhancing detection and classification accuracy. In this study, we introduce an innovative Attention-Aided Convolutional Neural Network (CNN) Autoencoder Network (AACNet) to improve automated diagnosis. AACNet employs a two-step methodology: first, a deep convolutional autoencoder with a soft attention gate learns and refines radiographic features in an unsupervised manner. Then, a CNN—built upon architectures like Xception, DenseNet-121, and ResNet-50—utilizes these enhanced feature maps for final classification. AACNet significantly surpasses traditional CNN models, addressing performance plateaus observed in conventional feature extraction. For instance, while DenseNet-121 alone achieves 91.56% accuracy, integrating it with AACNet boosts accuracy to 98.57%. Similarly, ResNet-50 and Xception paired with AACNet reach 97.74% and 98.25%, respectively. By mitigating overfitting and improving classification robustness, AACNet enhances AI-driven diagnostic precision. This advancement has profound implications for clinical practice, facilitating earlier detection and intervention, which are crucial for reducing mortality and healthcare burdens. Beyond COVID-19 and CAP, AACNet’s adaptability extends to diagnosing complex conditions such as cancer and brain tumors, underscoring its potential as a breakthrough in AI-powered medical imaging.