Development of an Eeg Artefact Detection Algorithm and its Application in Grading Hypoxic-Ischemic Encephalopathy

Objective: The purpose of this study is to develop a robust neonatal EEG artefact detection algorithm and to subsequently assess its application as a post-processing routine for automated EEG grading of hypoxic-ischemic encephalopathy (HIE). Methods: A database of neonatal EEG with expertly annotated artefacts was used to train and validate machine learning models to automatically identify EEG epochs containing artefacts. Approaches using a simple threshold-based digital signal processing (DSP) method, machine learning, and deep learning were developed and compared. The artefact detection classifier was subsequently assessed as a post-processing tool to assist in the application of automated EEG grading of HIE. An automated HIE grading algorithm was trained on a large dataset containing 653 hours of EEG. Post-processing routines were developed and assessed to integrate the artefact detection algorithm with the existing HIE grading algorithm. Results: Using a database containing 98 hours of EEG from 55 patients with per-channel annotations of artefacts, deep learning convolutional neural networks achieved best performance for artefact detection with 0.84 AUC, compared to 0.68 AUC and 0.82 AUC for a DSP method and a ridge classifier model, respectively. The developed algorithm provides a measure of quality for each 8-second epoch of EEG, which is a useful tool to assist in EEG interpretation. Using a database containing 653 hours of EEG from 181 patients, the automated HIE grading algorithm achieved 80.4% accuracy. The percentage of detected artefacts in the misclassified HIE epochs was not statistically higher than that of correctly classified HIE epochs. This demonstrates that the automated HIE grading algorithm is robust to artefacts. Despite this, incorporating a post-processing routine to automatically remove artefactual segments from the majority voting step of the HIE grading algorithm, achieved an improved performance of 81.7% accuracy. The database used in this study was originally pruned to be of high quality to begin with, and as such, real-world use of the HIE grading algorithm with artefact detection may yield greater improvements. Conclusion: Deep learning methods achieved highest classification performance for neonatal EEG artefact detection. The inclusion of artefact detection in automated HIE grading improves grading accuracy marginally and allows for quality measures to be presented alongside the automated HIE grades which may increase user confidence in its real-world application.