Scholarlink

<p>Histologic imagery is routinely used for the diagnosis of gastrointestinal (GI) diseases, but the manual examination is time-consuming and prone to inter-observer variability. This study presents a hybrid deep learning framework for automatic diagnosis and classification of GI disease states. The proposed approach combines a lightweight convolutional neural networks (CNN) and a traditional machine learning classifier in order to achieve a balance between the segmentation accuracy and the computational efficiency. CNN backbones such as ResNet18, MobileNet, GoogLeNet, ShuffleNet were used in combination with classifiers such as support vector machine (SVM), decision tree (DT), k-nearest neighbour (kNN). Each CNN is trained to learn deep-learning features from the histological images and a classifier is used to assign disease classes on the basis of deep-learning features. The effectiveness of the framework was tested in two histopathological datasets, MSI-MUT and MSS. The ResNet18-SVM combination resulted in the best accuracy (99%). Other ensemble combinations of CNN and classifiers also gave competitive results. The proposed architecture shows high accuracy, interpretability and computational efficiency making it appropriate for real-time or clinical deployment in the diagnosis of gastrointestinal disorders.</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

Transfer learning and hybrid deep convolutional neural networks for detection and classification of gastrointestinal diseases

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Ahmed F. Hussein,
  • Auns Q. Al-Neami,
  • Noor Kamal Al-Qazzaz

摘要

Histologic imagery is routinely used for the diagnosis of gastrointestinal (GI) diseases, but the manual examination is time-consuming and prone to inter-observer variability. This study presents a hybrid deep learning framework for automatic diagnosis and classification of GI disease states. The proposed approach combines a lightweight convolutional neural networks (CNN) and a traditional machine learning classifier in order to achieve a balance between the segmentation accuracy and the computational efficiency. CNN backbones such as ResNet18, MobileNet, GoogLeNet, ShuffleNet were used in combination with classifiers such as support vector machine (SVM), decision tree (DT), k-nearest neighbour (kNN). Each CNN is trained to learn deep-learning features from the histological images and a classifier is used to assign disease classes on the basis of deep-learning features. The effectiveness of the framework was tested in two histopathological datasets, MSI-MUT and MSS. The ResNet18-SVM combination resulted in the best accuracy (99%). Other ensemble combinations of CNN and classifiers also gave competitive results. The proposed architecture shows high accuracy, interpretability and computational efficiency making it appropriate for real-time or clinical deployment in the diagnosis of gastrointestinal disorders.