Scholarlink

<p>Respiratory illness remains a significant cause of increased morbidity and mortality rates, particularly in developing countries. To conduct an in-depth analysis an efficient machine-learning technique is essential. Various approaches including computed tomography (CT) have evolved in this domain to detect lung diseases but have several disadvantages such as the usage of ionization and inevitable radiation exposure by which the overall results get diminished. In response to this limitation, this paper proposes a novel approach called “Enhanced Multi-Scale Faster Recurrent Convolutional Neural Network based on Dove Swarm Optimization” (EMFRCNN-DSO). The DSO is an effective and time-efficient algorithm for solving several optimization problems. The proposed EMFRCNN-DSO algorithm addresses the demerits of several existing methods and demonstrates enhanced accuracy in classifying the classes of lung disease. The proposed EMFRCNN-DSO algorithm follows a structured process. It begins by preprocessing raw chest X-ray images through enhancement and normalization tasks. Subsequently, a set of significant features, including texture-based, shape-based, and histograms of oriented gradient features, are extracted from the preprocessed data. Based on these extracted feature vectors, the EMFRCNN-DSO classifier accurately categorizes multiple lung disease classes, including healthy, pneumonia, COPD, COVID-19, tuberculosis, and lung opacity. To enhance the classification performance further, the loss functions of the enhanced multi-scale Faster RCNN are normalized and weighted by tuning optimal values for the balancing parameter using the dove swarm optimization algorithm. This step significantly improves the accuracy of classifying lung disease classes. Experimental investigations validate the superiority of the proposed EMFRCNN-DSO algorithm, demonstrating an enhanced classification accuracy of approximately 97.5%. This achievement marks a notable improvement thus paving the way for more effective and timely interventions to combat respiratory illnesses.</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

An efficient lung disease classification using dove swarm optimization based multi-scale faster-RCNN model

  • 期刊论文

  • 发表日期：

    2025年7月12日

查看全文
  • Indumathi Varadharajan,
  • Siva Rathinavelayutham

摘要

Respiratory illness remains a significant cause of increased morbidity and mortality rates, particularly in developing countries. To conduct an in-depth analysis an efficient machine-learning technique is essential. Various approaches including computed tomography (CT) have evolved in this domain to detect lung diseases but have several disadvantages such as the usage of ionization and inevitable radiation exposure by which the overall results get diminished. In response to this limitation, this paper proposes a novel approach called “Enhanced Multi-Scale Faster Recurrent Convolutional Neural Network based on Dove Swarm Optimization” (EMFRCNN-DSO). The DSO is an effective and time-efficient algorithm for solving several optimization problems. The proposed EMFRCNN-DSO algorithm addresses the demerits of several existing methods and demonstrates enhanced accuracy in classifying the classes of lung disease. The proposed EMFRCNN-DSO algorithm follows a structured process. It begins by preprocessing raw chest X-ray images through enhancement and normalization tasks. Subsequently, a set of significant features, including texture-based, shape-based, and histograms of oriented gradient features, are extracted from the preprocessed data. Based on these extracted feature vectors, the EMFRCNN-DSO classifier accurately categorizes multiple lung disease classes, including healthy, pneumonia, COPD, COVID-19, tuberculosis, and lung opacity. To enhance the classification performance further, the loss functions of the enhanced multi-scale Faster RCNN are normalized and weighted by tuning optimal values for the balancing parameter using the dove swarm optimization algorithm. This step significantly improves the accuracy of classifying lung disease classes. Experimental investigations validate the superiority of the proposed EMFRCNN-DSO algorithm, demonstrating an enhanced classification accuracy of approximately 97.5%. This achievement marks a notable improvement thus paving the way for more effective and timely interventions to combat respiratory illnesses.