aiqbench analysis
This research proposes an attention-enhanced dual-branch neural network combining ConvNeXt and BiLSTM architectures to address the challenges of subject-independent EEG seizure detection. By jointly learning time-frequency and temporal representations, the model achieves high accuracy while maintaining low inference latency, demonstrating that specific architectural choices like temporal attention are critical for performance gains in this medical AI application.
Attention-Enhanced Dual-Branch ConvNeXt-BiLSTM Network for Subject-Independent EEG Seizure Detection
arXiv:2609.22141v1 Announce Type: cross Abstract: Automated seizure detection from scalp electroencephalography (EEG) is difficult because seizure morphology varies among patients and seizure samples are substantially outnumbered by non-seizure samples. This paper presents an attention-enhanced dual-branch network that jointly learns time--frequency and temporal representations from the same EEG segment. A continuous wavelet transform converts each segment into a scalogram processed by an ImageNet-pretrained ConvNeXt-Tiny backbone and squeeze-and-excitation attention. In parallel, a bidirectional long short-term memory network followed by multi-head self-attention models the raw signal. The two feature vectors are concatenated and classified by a weighted multilayer perceptron. Experiments use 14 subjects from the CHB-MIT scalp EEG database with subject-wise partitioning performed before overlapping segmentation. The model obtains $97.88\%$ accuracy and $97.51\%$ F1-score over ten across-subject splits, and $97.51\%$ accuracy, $96.59\%$ F1-score, and $98.02\%$ area under the ROC curve under 14-fold leave-one-subject-out validation. Removing temporal attention causes the largest ablation loss. The model requires 28.26 million parameters and 4.56 GFLOPs, with a measured network-only inference latency of 4.64 ms per segment.
Abstract:Automated seizure detection from scalp electroencephalography (EEG) is difficult because seizure morphology varies among patients and seizure samples are substantially outnumbered by non-seizure samples. This paper presents an attention-enhanced dual-branch network that jointly learns time--frequency and temporal representations from the same EEG segment. A continuous wavelet transform converts each segment into a scalogram processed by an ImageNet-pretrained ConvNeXt-Tiny backbone and squeeze-and-excitation attention. In parallel, a bidirectional long short-term memory network followed by multi-head self-attention models the raw signal. The two feature vectors are concatenated and classified by a weighted multilayer perceptron. Experiments use 14 subjects from the CHB-MIT scalp EEG database with subject-wise partitioning performed before overlapping segmentation. The model obtains $97.88\%$ accuracy and $97.51\%$ F1-score over ten across-subject splits, and $97.51\%$ accuracy, $96.59\%$ F1-score, and $98.02\%$ area under the ROC curve under 14-fold leave-one-subject-out validation. Removing temporal attention causes the largest ablation loss. The model requires 28.26 million parameters and 4.56 GFLOPs, with a measured network-only inference latency of 4.64 ms per segment.
Submitted to IEEE COMPAS 2026. 5 pages, 2 figures, 5 tables
Signal Processing (eess.SP); Machine Learning (cs.LG)
arXiv:2609.22141 [eess.SP]
(or arXiv:2609.22141v1 [eess.SP] for this version)
https://doi.org/10.48550/arXiv.2609.22141 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 270 words
Related coverage
Multi-encoder ConvNeXt Network with Smooth Attentional Feature Fusion for Multispectral Semantic Segmentation
arXiv · Oct 8, 2026
DAH-Net: A Dual-Attention Hybrid Network for Interpretable and Robust EEG-Based Emotion Recognition
arXiv · Jun 10, 2026
Digital States 2026: AI Moves From Experiment to Infrastructure
govtech.com · Sep 24, 2026
RAMamba-Net: A Reliability-Aware and Mamba-Based Multimodal Fusion Network for Auditory Attention Detection
arXiv · Sep 12, 2026