Window Stacking Meta-Models for Clinical EEG Classification

Yixuan Zhu,Rohan Kandasamy,Luke J. W. Canham,David Western

2024-01-15

Abstract:Windowing is a common technique in EEG machine learning classification and other time series tasks. However, a challenge arises when employing this technique: computational expense inhibits learning global relationships across an entire recording or set of recordings. Furthermore, the labels inherited by windows from their parent recordings may not accurately reflect the content of that window in isolation. To resolve these issues, we introduce a multi-stage model architecture, incorporating meta-learning principles tailored to time-windowed data aggregation. We further tested two distinct strategies to alleviate these issues: lengthening the window and utilizing overlapping to augment data. Our methods, when tested on the Temple University Hospital Abnormal EEG Corpus (TUAB), dramatically boosted the benchmark accuracy from 89.8 percent to 99.0 percent. This breakthrough performance surpasses prior performance projections for this dataset and paves the way for clinical applications of machine learning solutions to EEG interpretation challenges. On a broader and more varied dataset from the Temple University Hospital EEG Corpus (TUEG), we attained an accuracy of 86.7%, nearing the assumed performance ceiling set by variable inter-rater agreement on such datasets.

Signal Processing,Machine Learning

What problem does this paper attempt to address?

The paper focuses on the problems encountered in the classification of clinical EEG (Electroencephalogram), particularly the high computational cost and inaccurate labeling caused by the application of windowing techniques. To address these issues, the paper proposes a multi-stage model architecture combining meta-learning principles to handle aggregated time window data. The researchers also test two strategies, namely extending window length and incorporating overlap to increase data, in order to improve accuracy. On the Temple University Hospital Abnormal EEG Corpus (TUAB) dataset, this approach improves the baseline accuracy from 89.8% to 99.0%, representing a significant breakthrough in applying machine learning solutions to EEG interpretation challenges. On the broader and more diverse Temple University Hospital EEG Corpus (TUEG) dataset, they achieve an accuracy of 86.7%, nearing the performance upper limit of human expert consistency. The main contributions of the paper include optimizing the first-stage and second-stage model architectures to further improve accuracy on TUAB; evaluating the benefits of overlapping windows; extending the arbitration concept to session-level; evaluating on larger and more diverse datasets and using weighted loss functions to address sample imbalance; and studying the interpretability of the model. Through these methods, the paper demonstrates how performance in EEG classification tasks can be significantly improved by enhancing window handling and meta-model design, providing powerful tools for clinical applications.

Window Stacking Meta-Models for Clinical EEG Classification

Automatic Classification of Sleep Stages Based on Raw Single-Channel EEG

Windowing Compensation in Fourier Based Surrogate Analysis and Application to EEG Signal Classification

Electroencephalogram Emotion Recognition Based on A Stacking Classification Model

Improved Model Accuracy Based on a Simple Frame of Temporal-Correlation Representation Method in Sleep Staging

MSW-Transformer: Multi-Scale Shifted Windows Transformer Networks for 12-Lead ECG Classification

Improving Clinician Performance in Classification of EEG Patterns on the Ictal-Interictal-Injury Continuum using Interpretable Machine Learning

Performance investigation of epilepsy detection from noisy EEG signals using base-2-meta stacking classifier

Mixed Neural Network Approach for Temporal Sleep Stage Classification

Single-channel EEG sleep staging based on data augmentation and cross-subject discrepancy alleviation

Automatic Analysis of EEGs Using Big Data and Hybrid Deep Learning Architectures

Balancing Spectral, Temporal and Spatial Information for EEG-based Alzheimer's Disease Classification

An Interpretable Single-Channel EEG Sleep Staging Model Based on Prototype Matching and Multi-Task Learning

A Sliding Window Common Spatial Pattern for Enhancing Motor Imagery Classification in EEG-BCI

Improving Multi-Class Motor Imagery EEG Classification Using Overlapping Sliding Window and Deep Learning Model

Semi-automated Annotation of Signal Events in Clinical EEG Data

Representation learning for improved interpretability and classification accuracy of clinical factors from EEG

Automatic Detection of Abnormal EEG Signals Using WaveNet and LSTM

A Convolution Network of Multi-Windows Spatial-Temporal Feature Analysis for Single-trial EEG Classification in RSVP Task

Stabilizing Subject Transfer in EEG Classification with Divergence Estimation

Compact CNN with Dynamic Window for SSVEP-based BCIs