Scholarlink

Mycobacterium tuberculosis-caused tuberculosis (TB) continues to be one of the world’s most common infectious diseases, with significant rates of death and morbidity, especially in impoverished nations. Conventional diagnostic techniques such as culture tests and smear microscopy might take weeks to produce results. While bronchoscopy provides detailed imaging of the respiratory tract its diagnostic process is prone to variability and inefficiencies necessitating the development of automated and reliable diagnostic solutions. In order to improve TB identification from high-resolution bronchoscopy images, this paper suggests an ensemble deep learning (DL) methodology that combines Convolutional Neural Networks (CNNs) and ResNet architectures. CNN effectively captures spatial features such as edges and textures. ResNet models specialize in capturing deep hierarchical patterns using residual learning. The cross-entropy loss function which measures the disparity between predicted and actual probability distributions along with the Adam optimizer which adaptively adjusts learning rates for individual parameters are employed to train the model. This aids in minimizing overfitting and promotes stable training, particularly when handling complex datasets. With an accuracy of 97.62%, recall of 97.35%, precision of 97.74%, and an F1-score of 97.35%, the suggested method performs well and shows its dependability in correctly detecting TB patients. Additionally, a confusion matrix provides additional information about the model’s diagnostic performance that is significant for clinical applications where false negatives can delay treatment and false positives can impose psychological and medical burdens on patients. The implementation of this approach in MATLAB highlights its practical applicability given MATLAB extensive use in medical research for its robust toolboxes and advanced image processing capabilities. This paper represents a significant advancement in automated TB diagnostics enhancing accuracy while reducing interpretation variability in clinical settings.

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 Ensemble Approach for Tuberculosis Detection Using Bronchoscopy Diagnosis in Convolutional Neural Network and Residual Networks Through Feature Fusion

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • R. Santhosini,
  • P. Mohan Kumar,
  • B. Subha

摘要

Mycobacterium tuberculosis-caused tuberculosis (TB) continues to be one of the world’s most common infectious diseases, with significant rates of death and morbidity, especially in impoverished nations. Conventional diagnostic techniques such as culture tests and smear microscopy might take weeks to produce results. While bronchoscopy provides detailed imaging of the respiratory tract its diagnostic process is prone to variability and inefficiencies necessitating the development of automated and reliable diagnostic solutions. In order to improve TB identification from high-resolution bronchoscopy images, this paper suggests an ensemble deep learning (DL) methodology that combines Convolutional Neural Networks (CNNs) and ResNet architectures. CNN effectively captures spatial features such as edges and textures. ResNet models specialize in capturing deep hierarchical patterns using residual learning. The cross-entropy loss function which measures the disparity between predicted and actual probability distributions along with the Adam optimizer which adaptively adjusts learning rates for individual parameters are employed to train the model. This aids in minimizing overfitting and promotes stable training, particularly when handling complex datasets. With an accuracy of 97.62%, recall of 97.35%, precision of 97.74%, and an F1-score of 97.35%, the suggested method performs well and shows its dependability in correctly detecting TB patients. Additionally, a confusion matrix provides additional information about the model’s diagnostic performance that is significant for clinical applications where false negatives can delay treatment and false positives can impose psychological and medical burdens on patients. The implementation of this approach in MATLAB highlights its practical applicability given MATLAB extensive use in medical research for its robust toolboxes and advanced image processing capabilities. This paper represents a significant advancement in automated TB diagnostics enhancing accuracy while reducing interpretation variability in clinical settings.