MSCA-Net: A Multi-Scale Depthwise Attention Network for Multi-Class Intrusion Detection in Internet of Medical Things

Area of Science:

• Cybersecurity
• Network Security
• Healthcare Technology

Background:

• Internet of Medical Things (IoMT) networks are crucial for real-time healthcare but vulnerable to cyberattacks due to their complex nature and resource constraints.
• Existing attack detection systems often lack the low-latency, high-accuracy, and generalizability required for critical IoMT environments.

Purpose of the Study:

• To propose and evaluate the Multi-Scale Depthwise Channel Attention Network (MSCA-Net) for effective multi-class cyberattack detection in IoMT networks.
• To address challenges including multi-scale feature extraction, class imbalance, and computational efficiency in IoMT security.

Main Methods:

• Developed MSCA-Net, integrating multi-scale depthwise separable convolutions for temporal pattern capture, a channel attention mechanism for feature weighting, and a lightweight LSTM for temporal dependencies.
• Evaluated MSCA-Net on the WUSTL-EHMS-2020 and CICIoMT2024 datasets for multi-class and fine-grained attack classification.

Main Results:

• MSCA-Net achieved 99.75% accuracy and a 99.77% weighted F1 score in a 6-class scenario on the CICIoMT2024 dataset.
• Demonstrated competitive results in 19-class classification, outperforming nine baseline models in performance-to-cost ratio and offering up to 2x faster inference.
• Effectively handled multi-scale feature extraction, class imbalance, and computational efficiency.

Conclusions:

• The proposed MSCA-Net model provides a viable and efficient solution for real-time cyberattack detection in Internet of Medical Things environments.
• The model's architecture balances effective representation learning with low computational costs, making it suitable for resource-constrained IoMT systems.