Scholarlink

<p>Hospitals desire knowledge of bedside sensors in real time but they do not wish to send everything to the cloud. We propose an Edge-AI framework used in the IoT environment that maintains intelligence as near as possible to the patient, and ensures privacy and a trusted audit trail. A quantized CNN-LSTM is run on-device to detect anomalies and protects data in transit and at rest (TLS plus lightweight homomorphic encryption), logs notable events to a private block chain to give an auditable, tamper-proof history, and improves models off-device using federated learning to ensure that raw patient data never reaches the cloud. Reply The framework was tested against a cloud-only baseline on the MIT-BIH Arrhythmia dataset and a live multi-sensor synthetic stream, beating it across all metrics: accuracy 94.7% versus 83.1% accuracy, median inference latency 118&#xa0;ms versus 246&#xa0;ms (a 52% reduction), daily communication overhead 36.2&#xa0;MB versus 56.3&#xa0;MB (a savings of 38.1%), and energy usage 1.21 mW/sample (near 22% energy efficiency improvement). Block chain logging endured 75 events/s, which helped medico-legal tractability. Federated rounds provided ~ 1.5–2.3 accuracy points per round across five rounds, whereas INT8 quantization reduced model size by ~ 74% with only a ~ 0.4 accuracy degrade-practical in Jetson-class edge devices. SHAP explanations accompany alerts to enable the establishment of clinical trust in the model by demonstrating why a case was flagged. In general, the framework provides a safe, decipherable, and standard-aligned course to hospital-scale implementation (&gt; 1,000 nodes) with quantifiable gains in responsiveness and bandwidth as well as power.</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-AI enabled secure IoT framework for real-time patient monitoring and anomaly detection in smart healthcare systems

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月25日

查看全文
  • P. Karpagam,
  • M. Karthikeyan,
  • G. Kalpana,
  • A. Suresh

摘要

Hospitals desire knowledge of bedside sensors in real time but they do not wish to send everything to the cloud. We propose an Edge-AI framework used in the IoT environment that maintains intelligence as near as possible to the patient, and ensures privacy and a trusted audit trail. A quantized CNN-LSTM is run on-device to detect anomalies and protects data in transit and at rest (TLS plus lightweight homomorphic encryption), logs notable events to a private block chain to give an auditable, tamper-proof history, and improves models off-device using federated learning to ensure that raw patient data never reaches the cloud. Reply The framework was tested against a cloud-only baseline on the MIT-BIH Arrhythmia dataset and a live multi-sensor synthetic stream, beating it across all metrics: accuracy 94.7% versus 83.1% accuracy, median inference latency 118 ms versus 246 ms (a 52% reduction), daily communication overhead 36.2 MB versus 56.3 MB (a savings of 38.1%), and energy usage 1.21 mW/sample (near 22% energy efficiency improvement). Block chain logging endured 75 events/s, which helped medico-legal tractability. Federated rounds provided ~ 1.5–2.3 accuracy points per round across five rounds, whereas INT8 quantization reduced model size by ~ 74% with only a ~ 0.4 accuracy degrade-practical in Jetson-class edge devices. SHAP explanations accompany alerts to enable the establishment of clinical trust in the model by demonstrating why a case was flagged. In general, the framework provides a safe, decipherable, and standard-aligned course to hospital-scale implementation (> 1,000 nodes) with quantifiable gains in responsiveness and bandwidth as well as power.