A Lightweight Depthwise CNN with Efficient Attention Optimized by Chinese Pangolin Algorithm for Accurate and Low-latency IoMT intrusion Detection

The rapid adoption of Internet of Medical Things (IoMT) technologies has improved healthcare monitoring, diagnosis, and patient management while simultaneously increasing exposure to diverse cyber threats.This study proposes a Chinese Pangolin Optimization-based Lightweight Depthwise Convolutional Neural Network with Efficient Attention (CPOA-LDCNN-EA) for accurate and computationally efficient IoMT intrusion detection.The framework integrates depthwise separable convolution for lightweight feature extraction, Efficient Attention for discriminative feature enhancement, and Chinese Pangolin Optimization for joint feature-retention evaluation and hyperparameter optimization.CICIoMT2024 was employed as the source dataset for model development and internal evaluation, whereas ToN-IoT and CICIDS2017 were used exclusively as unseen target datasets for external crossdataset evaluation.On CICIoMT2024, the proposed framework achieved a five-run mean accuracy of 99.41 ± 0.18%, Macro-Precision of 99.27 ± 0.20%, Macro-Recall of 99.18 ± 0.23%, Macro-F1 of 99.22 ± 0.21%, and Macro PR-AUC of 0.9966 ± 0.0011.The representative median-performance run achieved approximately 99.40% Macro-Precision, 99.40% Macro-Recall, and 99.40% Macro-F1, with a Macro one-vs-rest ROC-AUC of 0.998 and a Macro PR-AUC of 0.997.Without target-domain retraining or optimization, the frozen source-trained model achieved accuracies of 98.87% on ToN-IoT and 98.54% on CICIDS2017, demonstrating high binary cross-dataset transfer accuracy under the adopted Benign/Attack harmonization protocol.The framework contains 3.2 million trainable parameters and requires 1.4 GFLOPs, with a model-forward latency of 1.58 ms/sample on an NVIDIA RTX 3080 GPU.Samecondition comparisons on CICIoMT2024 indicate a favorable accuracy-efficiency trade-off relative to the evaluated baselines.These findings demonstrate the potential of CPOA-LDCNN-EA for edge-assisted IoMT cybersecurity, while further evaluation on resource-constrained healthcare devices is required to establish direct on-device deployment performance.

Authors

Institutions

Publication Details

Journal
International journal of intelligent engineering and systems
Published
2026-11-17
DOI
https://doi.org/10.22266/ijies2026.1231.19
Primary Topic
Network Security and Intrusion Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Lightweight Depthwise CNN with Efficient Attention Optimized by Chinese Pangolin Algorithm for Accurate and Low-latency IoMT intrusion Detection

M P Sowbhagya, Renuka Vinod Chimankare, Valeti Nagarjuna, Senthil Kumar Jagatheesaperumal et al.
4 links
International journal of intelligent engineering and systems
Network Security and Intrusion Detection
article

A Lightweight Depthwise CNN with Efficient Attention Optimized by Chinese Pangolin Algorithm for Accurate and Low-latency IoMT intrusion Detection

M P Sowbhagya, Renuka Vinod Chimankare, Valeti Nagarjuna, Senthil Kumar Jagatheesaperumal, Jaishree R. Devaru, Nitesh N. Nikam, Ankita Avthankar, Potukuchi Ramadevi, Kavya D N, Prakash Gadipelli
article en

Abstract

The rapid adoption of Internet of Medical Things (IoMT) technologies has improved healthcare monitoring, diagnosis, and patient management while simultaneously increasing exposure to diverse cyber threats.This study proposes a Chinese Pangolin Optimization-based Lightweight Depthwise Convolutional Neural Network with Efficient Attention (CPOA-LDCNN-EA) for accurate and computationally efficient IoMT intrusion detection.The framework integrates depthwise separable convolution for lightweight feature extraction, Efficient Attention for discriminative feature enhancement, and Chinese Pangolin Optimization for joint feature-retention evaluation and hyperparameter optimization.CICIoMT2024 was employed as the source dataset for model development and internal evaluation, whereas ToN-IoT and CICIDS2017 were used exclusively as unseen target datasets for external crossdataset evaluation.On CICIoMT2024, the proposed framework achieved a five-run mean accuracy of 99.41 ± 0.18%, Macro-Precision of 99.27 ± 0.20%, Macro-Recall of 99.18 ± 0.23%, Macro-F1 of 99.22 ± 0.21%, and Macro PR-AUC of 0.9966 ± 0.0011.The representative median-performance run achieved approximately 99.40% Macro-Precision, 99.40% Macro-Recall, and 99.40% Macro-F1, with a Macro one-vs-rest ROC-AUC of 0.998 and a Macro PR-AUC of 0.997.Without target-domain retraining or optimization, the frozen source-trained model achieved accuracies of 98.87% on ToN-IoT and 98.54% on CICIDS2017, demonstrating high binary cross-dataset transfer accuracy under the adopted Benign/Attack harmonization protocol.The framework contains 3.2 million trainable parameters and requires 1.4 GFLOPs, with a model-forward latency of 1.58 ms/sample on an NVIDIA RTX 3080 GPU.Samecondition comparisons on CICIoMT2024 indicate a favorable accuracy-efficiency trade-off relative to the evaluated baselines.These findings demonstrate the potential of CPOA-LDCNN-EA for edge-assisted IoMT cybersecurity, while further evaluation on resource-constrained healthcare devices is required to establish direct on-device deployment performance.

International journal of intelligent engineering and systemsVol. 19(12)
University of Mumbai (IN), Symbiosis International University (IN), Dayananda Sagar University (IN), Dayananda Sagar College of Engineering (IN), CSMSS Chh. Shahu College of Engineering (IN), Aditya University (IN), Kammavari Sangham Institute of Technology (IN), Koneru Lakshmaiah Education Foundation (IN)
Network Security and Intrusion Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon