Scholarlink

<p>The proposed Model introduces an innovative real-time health monitoring system for detecting thyroid dysfunction and classifying thyroid nodules, integrating biomedical sensors and deep learning techniques. The system employs non-invasive sensors body temperature, pulse rate, SpO₂, galvanic skin response (GSR), and body pressure connected to an ESP controller to collect physiological data, which is transmitted to MATLAB for real-time analysis, enabling early detection of conditions like hyperthyroidism and hypothyroidism. A MATLAB App utilizes deep learning models bilateral CNN, wide CNN, and VGG16 to analyze this sensor data, while a transfer learning approach with convolutional neural network classifies thyroid nodules as benign or malignant from a dataset of 1,000–5,000 pre-processed ultrasound images. Data augmentation and a 70%-15%-15% training-validation-test split enhanced robustness, with performance assessed using metrics such as accuracy, precision, recall, and F1-score. This intelligent, non-invasive system provides a reliable decision-support tool with It attained 96% training accuracy and 94% validation accuracy for 50 epochs, with very little overfitting, outperforming traditional diagnostic methods and offering potential for broader medical imaging applications.</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

Smart Multimodal System for Early Diagnosis of Thyroid Dysfunction Using IoT Sensors and Deep Neural Networks

  • 期刊论文

  • 发表日期：

    2026年4月22日

查看全文
  • Akshatha Kamath,
  • Monica R. Mundada,
  • B. J. Sowmya,
  • S. Seema,
  • K. Kshreeraja

摘要

The proposed Model introduces an innovative real-time health monitoring system for detecting thyroid dysfunction and classifying thyroid nodules, integrating biomedical sensors and deep learning techniques. The system employs non-invasive sensors body temperature, pulse rate, SpO₂, galvanic skin response (GSR), and body pressure connected to an ESP controller to collect physiological data, which is transmitted to MATLAB for real-time analysis, enabling early detection of conditions like hyperthyroidism and hypothyroidism. A MATLAB App utilizes deep learning models bilateral CNN, wide CNN, and VGG16 to analyze this sensor data, while a transfer learning approach with convolutional neural network classifies thyroid nodules as benign or malignant from a dataset of 1,000–5,000 pre-processed ultrasound images. Data augmentation and a 70%-15%-15% training-validation-test split enhanced robustness, with performance assessed using metrics such as accuracy, precision, recall, and F1-score. This intelligent, non-invasive system provides a reliable decision-support tool with It attained 96% training accuracy and 94% validation accuracy for 50 epochs, with very little overfitting, outperforming traditional diagnostic methods and offering potential for broader medical imaging applications.