Scholarlink

<p>Liver fibrosis is a progressive liver disease that impairs liver function, necessitating a precise and timely diagnosis to ensure effective treatment and management. This paper describes a hybrid deep learning method for classifying liver fibrosis that utilises Convolutional Neural Networks (CNNs) in conjunction with an advanced Attention Mechanism to enhance performance in medical image analysis. Traditional methods for detecting liver fibrosis often suffer from issues such as overfitting, poor generalisation across different datasets, and difficulties in highlighting critical areas of interest in medical imaging. We propose a novel hybrid model that combines the EfficientNet-B7 architecture and a dual-attention mechanism to address these limitations. Incorporating spatial and channel attention mechanisms enhances EfficientNet-B7, a model renowned for its superior performance and computational efficiency. This integration enables the model to focus on the most crucial features of liver scans, thereby enhancing the accuracy of fibrosis detection and severity classification. We utilise multi-scale feature extraction with dilated convolutions to capture finer details, thereby enhancing model sensitivity. This approach introduces a significant innovation by combining EfficientNet-B7 with dual-attention mechanisms, which improves classification accuracy and computational efficiency while addressing common limitations in current solutions. The model is evaluated against the publicly available Liver Fibrosis Dataset (LiverFib), which comprises liver imaging scans annotated with corresponding fibrosis stages. Our hybrid model outperforms traditional CNN architectures, such as VGGNet, DenseNet, ResNet, and U-Net, as well as attention-based models, achieving 98.5% accuracy, 97.2% precision, 96.5% recall, 96.8% F1-score, and an AUC of 0.98. The attention mechanism enhances classification performance by focusing on critical areas of images, thereby increasing model interpretability and facilitating clinicians’ ability to identify fibrosis-related regions. The findings demonstrate that the proposed method is a highly effective tool for the automated detection and classification of liver fibrosis, providing high accuracy and valuable insights for informed clinical decision-making.</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

Advanced Liver Fibrosis Detection and Classification Through Deep Learning-Driven Image Analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月28日

查看全文
  • Umesh Kumar Lilhore,
  • Sarita Simaiya,
  • Roobaea Alroobaea,
  • Abdullah M. Baqasah,
  • Majed Alsafyani,
  • Afnan Alhazmi,
  • MD Monish Khan

摘要

Liver fibrosis is a progressive liver disease that impairs liver function, necessitating a precise and timely diagnosis to ensure effective treatment and management. This paper describes a hybrid deep learning method for classifying liver fibrosis that utilises Convolutional Neural Networks (CNNs) in conjunction with an advanced Attention Mechanism to enhance performance in medical image analysis. Traditional methods for detecting liver fibrosis often suffer from issues such as overfitting, poor generalisation across different datasets, and difficulties in highlighting critical areas of interest in medical imaging. We propose a novel hybrid model that combines the EfficientNet-B7 architecture and a dual-attention mechanism to address these limitations. Incorporating spatial and channel attention mechanisms enhances EfficientNet-B7, a model renowned for its superior performance and computational efficiency. This integration enables the model to focus on the most crucial features of liver scans, thereby enhancing the accuracy of fibrosis detection and severity classification. We utilise multi-scale feature extraction with dilated convolutions to capture finer details, thereby enhancing model sensitivity. This approach introduces a significant innovation by combining EfficientNet-B7 with dual-attention mechanisms, which improves classification accuracy and computational efficiency while addressing common limitations in current solutions. The model is evaluated against the publicly available Liver Fibrosis Dataset (LiverFib), which comprises liver imaging scans annotated with corresponding fibrosis stages. Our hybrid model outperforms traditional CNN architectures, such as VGGNet, DenseNet, ResNet, and U-Net, as well as attention-based models, achieving 98.5% accuracy, 97.2% precision, 96.5% recall, 96.8% F1-score, and an AUC of 0.98. The attention mechanism enhances classification performance by focusing on critical areas of images, thereby increasing model interpretability and facilitating clinicians’ ability to identify fibrosis-related regions. The findings demonstrate that the proposed method is a highly effective tool for the automated detection and classification of liver fibrosis, providing high accuracy and valuable insights for informed clinical decision-making.