Scholarlink

<p>In this paper, a novel framework called DeepFusionNet is presented. It is intended to accurately classify lung diseases using multi-modal imaging data, including CT, MRI, and X-ray images. Modern deep learning methods, such as TrifectaNet for reliable illness categorization and LungSegNet + + for precise lung segmentation, are integrated into the system. The suggested approach includes pre-processing, deep learning-based detection, feature extraction, feature selection, and identification of regions of interest. To be more precise, it uses median filtering to reduce noise, bilinear interpolation to resample, and Gamma Correlated Contrast Limited Adaptive Histogram Equalization (G-CLAHE) to normalize intensity across all modalities. In addition, the system makes use of modality-specific characteristics for thorough feature extraction, including statistical moments, Local Binary Patterns, edge detection, density histograms, and shape descriptors. The Hybrid Pufferfish-based Coati Optimization (HPCO) method is used for feature selection in order to find the most informative characteristics. Ultimately, SqueezeNet, Bidirectional Recurrent Neural Network (Bi-RNN), and Triplet Siamese Network models work together in a TrifectaNet ensemble model to perform classification, and Grad-CAM-based visualization is used to analyze and explain the classification choices. Compared to the current methods, the suggested model outperformed them, achieving 99.28% accuracy for 80% of the training set.</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

DEepfusionnet: Advanced Framework for Multi-modal Lung Disease Classification Using Lungsegnet + + and Trifectanet

  • 期刊论文

  • 发表日期：

    2025年11月17日

查看全文
  • Murine Sharmili S.,
  • Yesubai Rubavathi C.

摘要

In this paper, a novel framework called DeepFusionNet is presented. It is intended to accurately classify lung diseases using multi-modal imaging data, including CT, MRI, and X-ray images. Modern deep learning methods, such as TrifectaNet for reliable illness categorization and LungSegNet + + for precise lung segmentation, are integrated into the system. The suggested approach includes pre-processing, deep learning-based detection, feature extraction, feature selection, and identification of regions of interest. To be more precise, it uses median filtering to reduce noise, bilinear interpolation to resample, and Gamma Correlated Contrast Limited Adaptive Histogram Equalization (G-CLAHE) to normalize intensity across all modalities. In addition, the system makes use of modality-specific characteristics for thorough feature extraction, including statistical moments, Local Binary Patterns, edge detection, density histograms, and shape descriptors. The Hybrid Pufferfish-based Coati Optimization (HPCO) method is used for feature selection in order to find the most informative characteristics. Ultimately, SqueezeNet, Bidirectional Recurrent Neural Network (Bi-RNN), and Triplet Siamese Network models work together in a TrifectaNet ensemble model to perform classification, and Grad-CAM-based visualization is used to analyze and explain the classification choices. Compared to the current methods, the suggested model outperformed them, achieving 99.28% accuracy for 80% of the training set.