Image-based Data Representations of Time Series: A Comparative Analysis in EEG Artifact Detection

TL;DR

The paper investigates how image-based representations of time series can improve EEG artifact detection by evaluating six representations across eleven CNN architectures on the TUAR dataset. It introduces artifact detection in EEG as a controlled testbed, systematically profiling representation characteristics and model performance, and openly shares the testing framework to foster future benchmarking. The study finds that certain representations—particularly Markov Transition Fields, Correlation Matrices, and Spectrograms—strike a favorable balance between information preservation and noise suppression, with Xception and EfficientNetB0 often delivering top results. Overall, the work provides practical guidance on representation choice conditioned on signal quality and aims to catalyze broader, cross-domain comparative analyses through its open-source framework.

Abstract

Alternative data representations are powerful tools that augment the performance of downstream models. However, there is an abundance of such representations within the machine learning toolbox, and the field lacks a comparative understanding of the suitability of each representation method. In this paper, we propose artifact detection and classification within EEG data as a testbed for profiling image-based data representations of time series data. We then evaluate eleven popular deep learning architectures on each of six commonly-used representation methods. We find that, while the choice of representation entails a choice within the tradeoff between bias and variance, certain representations are practically more effective in highlighting features which increase the signal-to-noise ratio of the data. We present our results on EEG data, and open-source our testing framework to enable future comparative analyses in this vein.

Figures (7)

• Figure 1: Exemplar image representations of time series EEG data.
• Figure 2: Different artifacts in EEG data.
• Figure 3: Raw EEG data undergoes four preprocessing steps: (a) partitioning into 5s windows, (b) channel separation, normalization, and downsampling to 128 Hz, (c) creating visual representations for each channel, and (d) averaging them into a single image representing all channels (except for correlation matrices, which ensemble channels by default). For the purpose of illustration, we depict only 3 channels and 3 representation techniques.
• Figure 4: Accuracy and F1 scores on the validation set for each visualization technique, averaged across all models.
• Figure 5: Comparing the performance of different architectures on MTFs.