Preliminary Investigation of Visualizing Human Activity Recognition Neural Network

Calhoun

2022

Preprint

In recent years, the use of convolutional neural networks (CNNs) for raw electroencephalography (EEG) analysis has grown increasingly common. However, relative to earlier machine learning and deep learning methods with manually extracted features, CNNs for raw EEG analysis present unique problems for explainability. As such, a growing group of methods have been developed that provide insight into the spectral features learned by CNNs. However, spectral power is not the only important form of information within EEG, and the capacity to understand the roles of specific multispectral waveforms identified by CNNs could be very helpful. In this study, we present a novel model visualization-based approach that adapts the traditional CNN architecture to increase interpretability and combines that inherent interpretability with a systematic evaluation of the model via a series of novel explainability methods. Our approach evaluates the importance of spectrally distinct first-layer clusters of filters before examining the contributions of identified waveforms and spectra to cluster importance. We evaluate our approach within the context of automated sleep stage classification and find that, for the most part, our explainability results are highly consistent with clinical guidelines. Our approach is the first to systematically evaluate both waveform and spectral feature importance in CNNs trained on EEG data.

Section: Examining Importance Of Extracted Waveformsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A Systematic Approach for Explaining Time and Frequency Features Extracted by CNNs from Raw EEG Data

Calhoun

2022

Preprint

“…While the method does yield a sample in the time domain that maximizes activation for a particular class, it does not provide insight into the relative importance of different time domain features. A couple of other studies have used activation maximization for insight into the time domain ( Yoshimura et al, 2019 , 2021 ). However, they were only applied to networks trained on samples that were around 30 time points long, and EEG samples can be hundreds to thousands of time points long.…”

Section: Introductionmentioning

confidence: 99%

A Systematic Approach for Explaining Time and Frequency Features Extracted by Convolutional Neural Networks From Raw Electroencephalography Data

Calhoun

2022

Front. Neuroinform.

In recent years, the use of convolutional neural networks (CNNs) for raw resting-state electroencephalography (EEG) analysis has grown increasingly common. However, relative to earlier machine learning and deep learning methods with manually extracted features, CNNs for raw EEG analysis present unique problems for explainability. As such, a growing group of methods have been developed that provide insight into the spectral features learned by CNNs. However, spectral power is not the only important form of information within EEG, and the capacity to understand the roles of specific multispectral waveforms identified by CNNs could be very helpful. In this study, we present a novel model visualization-based approach that adapts the traditional CNN architecture to increase interpretability and combines that inherent interpretability with a systematic evaluation of the model via a series of novel explainability methods. Our approach evaluates the importance of spectrally distinct first-layer clusters of filters before examining the contributions of identified waveforms and spectra to cluster importance. We evaluate our approach within the context of automated sleep stage classification and find that, for the most part, our explainability results are highly consistent with clinical guidelines. Our approach is the first to systematically evaluate both waveform and spectral feature importance in CNNs trained on resting-state EEG data.

“…Activation maximization is a common prototyping approach that has mainly been used within the context of image analysis [19]. It has since been applied to timeseries analysis in a small number of instances [4], [20], [21]. In two of these instances, it was only used for very short time-series (i.e., approximately 30 points in length) within the domain of activity classification [20], [21].…”

Section: Introductionmentioning

confidence: 99%

“…It has since been applied to timeseries analysis in a small number of instances [4], [20], [21]. In two of these instances, it was only used for very short time-series (i.e., approximately 30 points in length) within the domain of activity classification [20], [21]. In the remaining instance, the authors input sinusoids into a model and examined the activation of the intermediate nodes of the model [4].…”

Section: Introductionmentioning

confidence: 99%

A Novel Activation Maximization-based Approach for Insight into Electrophysiology Classifiers

Sendi

et al. 2021

Preprint

Spectral analysis remains a hallmark approach for gaining insight into electrophysiology modalities like electroencephalography (EEG). As the field of deep learning has progressed, more studies have begun to train deep learning classifiers on raw EEG data, which presents unique problems for explainability. A growing number of studies have presented explainability approaches that provide insight into the spectral features learned by deep learning classifiers. However, existing approaches only attribute importance to different frequency bands. Most of the methods cannot provide insight into the actual spectral values or the relationship between spectral features that models have learned. Here, we present a novel adaptation of activation maximization for electrophysiology time-series that generates samples that indicate the features learned by classifiers by optimizing their spectral content. We evaluate our approach within the context of EEG sleep stage classification with a convolutional neural network, and we find that our approach is able to identify spectral patterns known to be associated with each sleep stage. We also find surprising results suggesting that our classifier may have prioritized the use of eye and motion artifact when identifying Awake samples. Our approach is the first adaptation of activation maximization to the domain of raw electrophysiology classification. Additionally, our approach has implications for explaining any classifier trained on highly dynamic, long time-series.