Scholarlink

The disease, Tuberculosis (TB) is a bacterial infection that results from Mycobacterium tuberculosis and can be fatal if not recognized early. Chest X-rays are used to identify TB, thus requiring close supervision from radiologists and doctors. Tuberculosis (TB) can be discovered and recognized through chest X-rays employing deep learning methodologies, leveraging the capabilities of Convolutional Neural Networks (CNNs) to differentiate between X-rays of individuals with TB and those that are normal. In this work, we developed a new deep learning methodology, named as TD-Net, incorporating pre-trained MobileNetV2 architecture with spatial and channel attention to elevate the accuracy, F1 Score, Recall and AUC-ROC of MobileNetV2. The experiment shows that the proposed TD-Net methodology gains a test accuracy of 98 and 97% respectively and an AUC-ROC score of 0.99 and 0.98 respectively on two benchmark TB datasets namely, a publicly available dataset curated by Twasifur Rahman et al. and TBX11 dataset.

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

TD-Net: A Deep Learning Technique with Spatial and Channel Attention for Efficient Tuberculosis Detection from Chest X-Ray Scans

  • 章节

  • 发表日期：

    2026年

查看全文
  • Debgandhar Ghosh,
  • Pawan Kumar Singh

摘要

The disease, Tuberculosis (TB) is a bacterial infection that results from Mycobacterium tuberculosis and can be fatal if not recognized early. Chest X-rays are used to identify TB, thus requiring close supervision from radiologists and doctors. Tuberculosis (TB) can be discovered and recognized through chest X-rays employing deep learning methodologies, leveraging the capabilities of Convolutional Neural Networks (CNNs) to differentiate between X-rays of individuals with TB and those that are normal. In this work, we developed a new deep learning methodology, named as TD-Net, incorporating pre-trained MobileNetV2 architecture with spatial and channel attention to elevate the accuracy, F1 Score, Recall and AUC-ROC of MobileNetV2. The experiment shows that the proposed TD-Net methodology gains a test accuracy of 98 and 97% respectively and an AUC-ROC score of 0.99 and 0.98 respectively on two benchmark TB datasets namely, a publicly available dataset curated by Twasifur Rahman et al. and TBX11 dataset.