Scholarlink

<p>Medical image (MI) classification is a vital component of computer-aided diagnosis (CAD) by enabling accurate detection of diseases across diverse imaging modalities. Existing medical image classification models often rely on single-path feature extraction and lack effective integration of heterogeneous features. Early and accurate classification of abnormalities in medical images is vital for timely diagnosis and clinical decision-making. To overcome these challenges, a novel MGFM-Net has been proposed for medical image classification into normal and abnormal categories. The proposed framework introduces a Dual Attention-Guided Multi-Branch GoogLeNet for extracting multi-scale spatial and semantic features by enabling accurate classification across diverse imaging modalities. Residual Group Attention (RGA) and Shuffle Group Attention (SGA) modules enhance feature representation by focusing on anatomically relevant regions. A Gated Modality Fusion Layer integrates multiple modality-specific derived features to produce a discriminative latent representation. The final classification layer generates a binary output (normal vs. abnormal) thereby enabling accurate and reliable computer-aided diagnosis. The proposed MGFM-Net was evaluated using the specified measures like accuracy (AY), specificity (SY), recall (RL), precision (PN) and F1 score (FS). As a result of the experimental analysis, the proposed MGFM-Net achieves an overall AY of 99.04% and an F1 score of 98.58%. The proposed MGFM-Net enhances overall accuracy by 19.33%, 11.21%, 8.71%, 5.64%, 0.26%, 1.06% and 0.79% compared to HiFuse, ResNet50, Embedded Accelerated Systems, and DDCNN, MFFDCNN-CTDC, Federated Vision Transformer, Visual State Space (VSS), respectively.</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

MGFM-Net: Dual Attention-Guided Multi-Branch Network with Gated Feature Fusion for Medical Image Classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Jeyalakshmi Subbiah,
  • Kirubanandasarathy Nageswaran

摘要

Medical image (MI) classification is a vital component of computer-aided diagnosis (CAD) by enabling accurate detection of diseases across diverse imaging modalities. Existing medical image classification models often rely on single-path feature extraction and lack effective integration of heterogeneous features. Early and accurate classification of abnormalities in medical images is vital for timely diagnosis and clinical decision-making. To overcome these challenges, a novel MGFM-Net has been proposed for medical image classification into normal and abnormal categories. The proposed framework introduces a Dual Attention-Guided Multi-Branch GoogLeNet for extracting multi-scale spatial and semantic features by enabling accurate classification across diverse imaging modalities. Residual Group Attention (RGA) and Shuffle Group Attention (SGA) modules enhance feature representation by focusing on anatomically relevant regions. A Gated Modality Fusion Layer integrates multiple modality-specific derived features to produce a discriminative latent representation. The final classification layer generates a binary output (normal vs. abnormal) thereby enabling accurate and reliable computer-aided diagnosis. The proposed MGFM-Net was evaluated using the specified measures like accuracy (AY), specificity (SY), recall (RL), precision (PN) and F1 score (FS). As a result of the experimental analysis, the proposed MGFM-Net achieves an overall AY of 99.04% and an F1 score of 98.58%. The proposed MGFM-Net enhances overall accuracy by 19.33%, 11.21%, 8.71%, 5.64%, 0.26%, 1.06% and 0.79% compared to HiFuse, ResNet50, Embedded Accelerated Systems, and DDCNN, MFFDCNN-CTDC, Federated Vision Transformer, Visual State Space (VSS), respectively.