Scholarlink

Purpose <p>The primary purpose of this study is to investigate whether low-cost thermal imaging, combined with optimized deep learning techniques, can provide an accurate, safe, and non-invasive solution for the automated diagnosis and therapeutic monitoring of respiratory diseases. Specifically, the research addresses the question of whether thermal patterns associated with lung inflammation can be effectively learned and discriminated to reliably classify COVID-19, pneumonia, and healthy cases, while maintaining high diagnostic accuracy and model interpretability suitable for resource-limited clinical settings.</p> Methods <p>A thermal imaging–driven diagnostic framework was designed incorporating robust image preprocessing, multi-domain handcrafted feature extraction, and a hybrid feature selection strategy that combines the Coati Optimization Algorithm and the Marine Predators Algorithm. Customized deep learning models, including multilayer perceptron (MLP), recurrent neural network (RNN), attention-based convolutional neural network (CNN), and transfer learning–based architectures, were trained and evaluated using simulated thermal images representing COVID-19, pneumonia, and healthy conditions. Model interpretability was enhanced using Gradient-weighted Class Activation Mapping (Grad-CAM).</p> Results <p>The optimized framework achieved high classification performance, with accuracies of 98.58% using a 70/30 train–test split and 99.57% using an 80/20 split. Grad-CAM visualizations consistently identified disease-relevant thermal regions that contributed to model predictions, supporting both diagnostic reliability and interpretability.Conclusion:The proposed low-cost thermal imaging and deep learning framework demonstrates strong potential as an accessible, radiation-free solution for automated screening and monitoring of respiratory diseases. The high diagnostic accuracy and interpretability indicate its suitability for large-scale, resource-constrained healthcare settings.</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

Therapeutic Monitoring and Non-Invasive Diagnosis of Lung Disease (Pneumonia and COVID-19) Using Low-Cost Thermal Imaging and DEEP Learning

  • 期刊论文

  • 发表日期：

    2026年1月12日

查看全文
  • Murine Sharmili S,
  • Yesubai Rubavathy C

摘要

Purpose

The primary purpose of this study is to investigate whether low-cost thermal imaging, combined with optimized deep learning techniques, can provide an accurate, safe, and non-invasive solution for the automated diagnosis and therapeutic monitoring of respiratory diseases. Specifically, the research addresses the question of whether thermal patterns associated with lung inflammation can be effectively learned and discriminated to reliably classify COVID-19, pneumonia, and healthy cases, while maintaining high diagnostic accuracy and model interpretability suitable for resource-limited clinical settings.

Methods

A thermal imaging–driven diagnostic framework was designed incorporating robust image preprocessing, multi-domain handcrafted feature extraction, and a hybrid feature selection strategy that combines the Coati Optimization Algorithm and the Marine Predators Algorithm. Customized deep learning models, including multilayer perceptron (MLP), recurrent neural network (RNN), attention-based convolutional neural network (CNN), and transfer learning–based architectures, were trained and evaluated using simulated thermal images representing COVID-19, pneumonia, and healthy conditions. Model interpretability was enhanced using Gradient-weighted Class Activation Mapping (Grad-CAM).

Results

The optimized framework achieved high classification performance, with accuracies of 98.58% using a 70/30 train–test split and 99.57% using an 80/20 split. Grad-CAM visualizations consistently identified disease-relevant thermal regions that contributed to model predictions, supporting both diagnostic reliability and interpretability.Conclusion:The proposed low-cost thermal imaging and deep learning framework demonstrates strong potential as an accessible, radiation-free solution for automated screening and monitoring of respiratory diseases. The high diagnostic accuracy and interpretability indicate its suitability for large-scale, resource-constrained healthcare settings.