Scholarlink

<p>This study develops an enhanced GoogLeNet model integrated with an intuitionistic fuzzy Gaussian membership function preprocessing mechanism (GoogLeNet_IF) for classifying COVID-19 computed tomography (CT) images. The approach introduces a novel combination of Intuitionistic Fuzzy Sets (IFs) and Gaussian membership functions. It enables adaptive adjustment of pixel-level membership degrees to achieve more accurate and discriminative image feature representation. The intuitionistic fuzzy Gaussian-based preprocessing framework enhances the clarity and detail of medical images before feature extraction. By synergistically integrating the powerful feature learning capability of GoogLeNet with the robustness to uncertainty of IFs, the proposed GoogLeNet_IF model exhibits superior classification robustness. Particle Swarm Optimization (PSO) is employed to jointly optimize intuitionistic fuzzy parameters and the GoogLeNet hyperparameters while ensuring optimal model performance. Six variants are evaluated: (1) the baseline GoogLeNet, (2) a fuzzy Gaussian—enhanced GoogLeNet, (3) an intuitionistic fuzzy GoogLeNet with fuzzy parameter optimization, (4) a fully optimized GoogLeNet_IF model, (5) U-Net, and (6) ResNet. Experiments on the COVID-19 CT datasets demonstrate that the proposed models achieve remarkable performance, with mean accuracies of 99.20%, 92.53%, and 98.25%, corresponding standard deviations of 0.0034, 0.0210, and 0.0120 for Cases I–III, respectively. Statistical analyses using F tests and two-sample <i>t</i> tests confirm significant improvements in accuracy and stability. The results indicate that advanced optimization algorithms combined with intuitionistic fuzzy logic can improve the reliability and accuracy of AI medical imaging for COVID-19 detection and related clinical use.</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

Improved GoogLeNet with Intuitionistic Fuzzy Gaussian Membership Function Preprocessing for CT Images

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Kuo-Ping Lin,
  • Ting-Yu Lin,
  • Chieh-An Liu,
  • Ding-Hsiang Huang,
  • Yu-Tse Tsan

摘要

This study develops an enhanced GoogLeNet model integrated with an intuitionistic fuzzy Gaussian membership function preprocessing mechanism (GoogLeNet_IF) for classifying COVID-19 computed tomography (CT) images. The approach introduces a novel combination of Intuitionistic Fuzzy Sets (IFs) and Gaussian membership functions. It enables adaptive adjustment of pixel-level membership degrees to achieve more accurate and discriminative image feature representation. The intuitionistic fuzzy Gaussian-based preprocessing framework enhances the clarity and detail of medical images before feature extraction. By synergistically integrating the powerful feature learning capability of GoogLeNet with the robustness to uncertainty of IFs, the proposed GoogLeNet_IF model exhibits superior classification robustness. Particle Swarm Optimization (PSO) is employed to jointly optimize intuitionistic fuzzy parameters and the GoogLeNet hyperparameters while ensuring optimal model performance. Six variants are evaluated: (1) the baseline GoogLeNet, (2) a fuzzy Gaussian—enhanced GoogLeNet, (3) an intuitionistic fuzzy GoogLeNet with fuzzy parameter optimization, (4) a fully optimized GoogLeNet_IF model, (5) U-Net, and (6) ResNet. Experiments on the COVID-19 CT datasets demonstrate that the proposed models achieve remarkable performance, with mean accuracies of 99.20%, 92.53%, and 98.25%, corresponding standard deviations of 0.0034, 0.0210, and 0.0120 for Cases I–III, respectively. Statistical analyses using F tests and two-sample t tests confirm significant improvements in accuracy and stability. The results indicate that advanced optimization algorithms combined with intuitionistic fuzzy logic can improve the reliability and accuracy of AI medical imaging for COVID-19 detection and related clinical use.