Scholarlink

<p>The timely identification of COVID-19 is essential to mitigate elevated mortality rates and prevent the future proliferation of the pandemic. Diagnostic test kits identify the illness; however, they often require considerable time and may present challenges regarding accuracy. Chest computed tomography (CT) testing demonstrates greater accuracy and can also serve as a diagnostic tool for COVID-19. This empirical investigation utilizes a three-step innovative and highly effective hybrid methodology that integrates image processing, soft computing, and machine learning techniques to detect COVID-19 from chest CT scans. Following the pre-processing of the chest CT images, we proceeded to extract 213 features categorized into various classes. During the second phase, the selection of the most informative features essential for prediction was conducted utilizing three soft computing algorithms: the grey wolf optimization algorithm (GWO), the salp swarm-based optimization algorithm (SSA), and a proposed hybrid approach combining both algorithms. Leveraging the features identified through soft computing algorithms, the five benchmark machine learning models were trained and utilized to classify these CT images into COVID-19 and non-COVID categories based on the chosen feature set. We undertook comprehensive testing involving 24 distinct tests (utilizing both 5-fold and 10-fold cross-validation approach). For each test, we evaluated performance across nine different standard metrics. The proposed model underwent evaluation using a publicly available CT dataset, achieving an impressive AUC of 0.9999 and a notable maximum accuracy rate of 96.90%. The findings from the experiments indicate that the suggested prediction approach surpasses other contemporary leading techniques. Expert radiologists facing heavy workloads may find the results of the proposed medical decision support framework beneficial as an additional perspective. The proposed solution is also particularly beneficial in areas experiencing a dearth of experienced medical practitioners. Additionally, we expect the suggested clinical decision system to be generalizable, and by choosing the most influential features for predictions of other human diseases, the same performance might be replicated on other datasets of human diseases.</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

A hybrid feature selection framework collaborating image processing, evolutionary intelligence, and machine learning for covid-19 disease classification

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Law Kumar Singh,
  • Munish Khanna,
  • Amit Yadav,
  • Rekha singh

摘要

The timely identification of COVID-19 is essential to mitigate elevated mortality rates and prevent the future proliferation of the pandemic. Diagnostic test kits identify the illness; however, they often require considerable time and may present challenges regarding accuracy. Chest computed tomography (CT) testing demonstrates greater accuracy and can also serve as a diagnostic tool for COVID-19. This empirical investigation utilizes a three-step innovative and highly effective hybrid methodology that integrates image processing, soft computing, and machine learning techniques to detect COVID-19 from chest CT scans. Following the pre-processing of the chest CT images, we proceeded to extract 213 features categorized into various classes. During the second phase, the selection of the most informative features essential for prediction was conducted utilizing three soft computing algorithms: the grey wolf optimization algorithm (GWO), the salp swarm-based optimization algorithm (SSA), and a proposed hybrid approach combining both algorithms. Leveraging the features identified through soft computing algorithms, the five benchmark machine learning models were trained and utilized to classify these CT images into COVID-19 and non-COVID categories based on the chosen feature set. We undertook comprehensive testing involving 24 distinct tests (utilizing both 5-fold and 10-fold cross-validation approach). For each test, we evaluated performance across nine different standard metrics. The proposed model underwent evaluation using a publicly available CT dataset, achieving an impressive AUC of 0.9999 and a notable maximum accuracy rate of 96.90%. The findings from the experiments indicate that the suggested prediction approach surpasses other contemporary leading techniques. Expert radiologists facing heavy workloads may find the results of the proposed medical decision support framework beneficial as an additional perspective. The proposed solution is also particularly beneficial in areas experiencing a dearth of experienced medical practitioners. Additionally, we expect the suggested clinical decision system to be generalizable, and by choosing the most influential features for predictions of other human diseases, the same performance might be replicated on other datasets of human diseases.