Scholarlink

<p>This paper proposes a novel content-based medical image retrieval and classification model that combines handcrafted feature descriptors, metaheuristic optimization, and neural network classification. The Local Diagonal Extrema Pattern (LDEP) and Local Diagonal Laplacian Pattern (LDLP) descriptors are fused to capture complementary structural and texture features from medical images across various modalities. The fused features are optimized using Moth-Flame Optimization (MFO) to reduce dimensionality and enhance discriminative efficiency. A lightweight feedforward Neural Network (NN) is then used for classification, while the Canberra distance metric is employed for retrieval due to its sensitivity to sparse and low-valued features. The proposed model is implemented in MATLAB and evaluated on multiple publicly available medical imaging datasets. Finally, different descriptors and optimization methods are used to evaluate the model's effectiveness. The precision of the proposed retrieval model is 15.6%, 12.34%, 11.43%, 10.58%, and 7.5% higher than LBP, LDP, LTP, LDEP, and LDLP, respectively. Similarly, the precision of the proposed retrieval model is 14.3%, 11.57%, 10.26%, and 5.8% higher compared to other optimization models, demonstrating its superior performance in retrieval accuracy.</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

Fusion of optimized feature descriptors for medical image retrieval and classification

  • 期刊论文

  • 发表日期：

    2025年10月1日

查看全文
  • K. Revathi,
  • S. Vijaya Kumar

摘要

This paper proposes a novel content-based medical image retrieval and classification model that combines handcrafted feature descriptors, metaheuristic optimization, and neural network classification. The Local Diagonal Extrema Pattern (LDEP) and Local Diagonal Laplacian Pattern (LDLP) descriptors are fused to capture complementary structural and texture features from medical images across various modalities. The fused features are optimized using Moth-Flame Optimization (MFO) to reduce dimensionality and enhance discriminative efficiency. A lightweight feedforward Neural Network (NN) is then used for classification, while the Canberra distance metric is employed for retrieval due to its sensitivity to sparse and low-valued features. The proposed model is implemented in MATLAB and evaluated on multiple publicly available medical imaging datasets. Finally, different descriptors and optimization methods are used to evaluate the model's effectiveness. The precision of the proposed retrieval model is 15.6%, 12.34%, 11.43%, 10.58%, and 7.5% higher than LBP, LDP, LTP, LDEP, and LDLP, respectively. Similarly, the precision of the proposed retrieval model is 14.3%, 11.57%, 10.26%, and 5.8% higher compared to other optimization models, demonstrating its superior performance in retrieval accuracy.