Scholarlink

<p>Tuberculosis (TB) is currently recognized as a bacterial infection caused by Mycobacterium tuberculosis (MTB) that mainly targets the lungs. Recently, various efforts have been made to develop automated techniques for detecting TB bacteria in microscopic sputum smear images. The sputum image is the most employed diagnostic modality to identify pulmonary tuberculosis. These automatic methods are generally regarded as effective solutions to support the diagnostic process. People with pulmonary tuberculosis may not exhibit symptoms until the disease reaches advanced stages, making early detection highly challenging. To improve the detection rate, this research develops an effective approach named Fractal Kronecker Forward Fractional Network (FKFFNet) for detecting the severity of tuberculosis at its initial stages using sputum images. At first, the sputum image is pre-processed using the Kuwahara Filters, and then, the Bacilli Segmentation is done based on SegNet. After the segmentation, the extraction of features is done based on Speeded-Up Robust Features (SURF), statistical features, and Gray Level Co-occurrence Matrix (GLCM). Next, detection of tuberculosis severity levels is performed using FKFFNet, which integrates FractalNet and Deep Kronecker Network (DKN) through a fractional concept. The fractional concept applies fractional-order weighting during the feature fusion to enhance generalization and mitigate the overfitting issue. Additionally, the performance of FKFFNet is assessed using different metrics and obtained maximum values of 90.393% for accuracy, 89.512% for True Positive Rate (TPR), 91.730% for True Negative Rate (TNR), 88.109% for Positive Predictive Value (PPV), and 84.839% for Negative Predictive Value (NPV). When considering the accuracy metrics, the performance improvement gained by the devised FKFFNet model is 9.832%, 8.095%, 6.132%, 4.428%, 5.887%, and 2.830% higher than the Tuberculosis Severity Level Categorizing Algorithm (TSLCA), CTBViT, Generative Adversarial Network (GAN), TB-CXR-Net, AlexNet, and K-Nearest Neighbor (KNN).</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

FKFFNet: Fractal Kronecker Forward Fractional Net for Severity Detection of Tuberculosis Using Sputum Image

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Sanjeevkumar Angadi,
  • Saili Hemant Sable,
  • Shital Bhushan Mehta,
  • Rajani Amol Hemade,
  • Vaibhavi Umesh Avachat

摘要

Tuberculosis (TB) is currently recognized as a bacterial infection caused by Mycobacterium tuberculosis (MTB) that mainly targets the lungs. Recently, various efforts have been made to develop automated techniques for detecting TB bacteria in microscopic sputum smear images. The sputum image is the most employed diagnostic modality to identify pulmonary tuberculosis. These automatic methods are generally regarded as effective solutions to support the diagnostic process. People with pulmonary tuberculosis may not exhibit symptoms until the disease reaches advanced stages, making early detection highly challenging. To improve the detection rate, this research develops an effective approach named Fractal Kronecker Forward Fractional Network (FKFFNet) for detecting the severity of tuberculosis at its initial stages using sputum images. At first, the sputum image is pre-processed using the Kuwahara Filters, and then, the Bacilli Segmentation is done based on SegNet. After the segmentation, the extraction of features is done based on Speeded-Up Robust Features (SURF), statistical features, and Gray Level Co-occurrence Matrix (GLCM). Next, detection of tuberculosis severity levels is performed using FKFFNet, which integrates FractalNet and Deep Kronecker Network (DKN) through a fractional concept. The fractional concept applies fractional-order weighting during the feature fusion to enhance generalization and mitigate the overfitting issue. Additionally, the performance of FKFFNet is assessed using different metrics and obtained maximum values of 90.393% for accuracy, 89.512% for True Positive Rate (TPR), 91.730% for True Negative Rate (TNR), 88.109% for Positive Predictive Value (PPV), and 84.839% for Negative Predictive Value (NPV). When considering the accuracy metrics, the performance improvement gained by the devised FKFFNet model is 9.832%, 8.095%, 6.132%, 4.428%, 5.887%, and 2.830% higher than the Tuberculosis Severity Level Categorizing Algorithm (TSLCA), CTBViT, Generative Adversarial Network (GAN), TB-CXR-Net, AlexNet, and K-Nearest Neighbor (KNN).