Scholarlink

Purpose <p>Critical and precise discrimination between bacterial and viral pneumonia in pediatric patients is necessary since their treatment strategies are different. A real-time embedded system for pediatric pneumonia diagnosis from chest X-ray images using deep learning with built-in remote monitoring is discussed in this paper.</p> Methods <p>Five lightweight convolutional neural networks (CNNs)—ResNet-18, MobileNet, ShuffleNet, SqueezeNet, and EfficientNet-B0—were evaluated on a set of 5216 pediatric chest X-rays consisting of 2530 bacterial, 1345 viral, and 1341 normal. The best-performing model was optimized with NVIDIA TensorRT and deployed on a Jetson Nano for real-time edge inference. Diagnosis results were transmitted using the MQTT protocol to the ThingSpeak cloud platform for remote monitoring.</p> Results <p>MobileNet achieved the highest F1-score (95.7%), followed closely by EfficientNet-B0 (95.1%) and ResNet-18 (94.4%). MobileNet was selected based on efficiency and accuracy. The optimized implementation demonstrated high-accuracy, low-latency inference suitable for embedded environments.</p> Conclusion <p>The suggested system demonstrates an effective and resource-efficient approach to pediatric pneumonia diagnosis. By combining lightweight CNN and remote monitoring, the system offers a practical and scalable solution for accurate and rapid pneumonia detection.</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

Edge-based real-time diagnosis of pediatric pneumonia using lightweight CNNs and chest X-rays

  • 期刊论文

  • 发表日期：

    2025年11月14日

查看全文
  • Elmehdi Benmalek,
  • Wajih Rhalem,
  • Atman Jbabi,
  • Abdelilah Jilbab,
  • Jamal Elmhamdi

摘要

Purpose

Critical and precise discrimination between bacterial and viral pneumonia in pediatric patients is necessary since their treatment strategies are different. A real-time embedded system for pediatric pneumonia diagnosis from chest X-ray images using deep learning with built-in remote monitoring is discussed in this paper.

Methods

Five lightweight convolutional neural networks (CNNs)—ResNet-18, MobileNet, ShuffleNet, SqueezeNet, and EfficientNet-B0—were evaluated on a set of 5216 pediatric chest X-rays consisting of 2530 bacterial, 1345 viral, and 1341 normal. The best-performing model was optimized with NVIDIA TensorRT and deployed on a Jetson Nano for real-time edge inference. Diagnosis results were transmitted using the MQTT protocol to the ThingSpeak cloud platform for remote monitoring.

Results

MobileNet achieved the highest F1-score (95.7%), followed closely by EfficientNet-B0 (95.1%) and ResNet-18 (94.4%). MobileNet was selected based on efficiency and accuracy. The optimized implementation demonstrated high-accuracy, low-latency inference suitable for embedded environments.

Conclusion

The suggested system demonstrates an effective and resource-efficient approach to pediatric pneumonia diagnosis. By combining lightweight CNN and remote monitoring, the system offers a practical and scalable solution for accurate and rapid pneumonia detection.