Scholarlink

<p>Remote diagnoses and real-time monitoring, facilitated by the spread of medical Internet of Things (IoMT) devices, have revolutionized patient care. IoMT infrastructure expands when expanded, creating a bigger attack surface and exposing it to cyberattacks that threaten patient safety and data integrity. This paper proposes a lightweight Transformer-based framework, optimized for medical IoT, for intrusion detection and classification. The common limitations are identified using IoMT-TrafficData and the IoT Healthcare Security Dataset. Packet- and flow-level features are combined using depth-wise separable convolutions and a two-layer Transformer encoder. Class imbalance is addressed through the use of SMOTE paired with cost-sensitive learning. The results of experimental evaluation show up to 97.9% F1-score in multiclass classification, which performs better than CNN-LSTM hybrids and decreases inference latency by 25% on representative edge hardware. By finding the most helpful features, attention-based explainability boosts trust and reveals actionable insights. This system features innovative, real-time, explainable, low-latency attack detection for medical IoT, improving connected healthcare system security and resilience.</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

Edgeguardmed: Transformer-Based Intrusion Detection with Packet-Flow Fusion for Resilient IoMT

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • E. Bhuvaneswari,
  • G. Uthradevi,
  • A. Mary Joy Kinol,
  • P. Shanthi

摘要

Remote diagnoses and real-time monitoring, facilitated by the spread of medical Internet of Things (IoMT) devices, have revolutionized patient care. IoMT infrastructure expands when expanded, creating a bigger attack surface and exposing it to cyberattacks that threaten patient safety and data integrity. This paper proposes a lightweight Transformer-based framework, optimized for medical IoT, for intrusion detection and classification. The common limitations are identified using IoMT-TrafficData and the IoT Healthcare Security Dataset. Packet- and flow-level features are combined using depth-wise separable convolutions and a two-layer Transformer encoder. Class imbalance is addressed through the use of SMOTE paired with cost-sensitive learning. The results of experimental evaluation show up to 97.9% F1-score in multiclass classification, which performs better than CNN-LSTM hybrids and decreases inference latency by 25% on representative edge hardware. By finding the most helpful features, attention-based explainability boosts trust and reveals actionable insights. This system features innovative, real-time, explainable, low-latency attack detection for medical IoT, improving connected healthcare system security and resilience.