vEpiNet: A multimodal interictal epileptiform discharge detection method based on video and electroencephalogram...

Area of Science:

• Medical technology
• Artificial intelligence
• Neuroscience

Background:

• Automated detection of interictal epileptiform discharges (IEDs) is crucial for epilepsy diagnosis and treatment.
• Current deep learning methods primarily rely on electroencephalogram (EEG) data, potentially missing valuable information from other modalities.
• Enhancing the accuracy and efficiency of IED detection systems is an ongoing challenge in clinical practice.

Purpose of the Study:

• To develop and evaluate a multimodal deep learning model, vEpiNet, that integrates video and EEG data for enhanced automated IED detection.
• To assess the contribution of video data to IED detection performance compared to an EEG-only model (nEpiNet).
• To validate the clinical viability and real-world performance of the vEpiNet system.

Main Methods:

• A multimodal deep learning framework, vEpiNet, was proposed, processing both video (using frame difference and Simple Keypoints for movement) and EEG (using EfficientNetV2) data.
• Video and EEG features were fused using a multilayer perceptron.
• A comparative model, nEpiNet (EEG-only), was developed to isolate the impact of video data.
• Model performance was evaluated using 10-fold cross-validation and a prospective real-world clinical dataset.

Main Results:

• vEpiNet achieved a superior area under the receiver operating characteristic curve (AUROC) of 0.9902 compared to nEpiNet's 0.9878 in cross-validation.
• In a prospective test dataset, vEpiNet demonstrated a higher area under the precision-recall curve (AUPRC) of 0.8623 versus nEpiNet's 0.8316.
• Incorporating video data increased precision to 76.6% (at 80% sensitivity) and reduced false positives by nearly one-third, with efficient processing times (5.7 min/hour).

Conclusions:

• Multimodal integration of video and EEG data significantly improves the performance and precision of automated IED detection.
• The vEpiNet model shows strong potential for clinical application, offering a viable and effective tool for real-world IED analysis.
• Video data provides complementary information that enhances the robustness and accuracy of deep learning-based epilepsy detection systems.