Biology subjects

Ellis, C. A.

Publications and source records attributed to Ellis, C. A..

12 recordsLinked to original sources

bioRxiv2022

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

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.

bioinformatics↗

bioRxiv2022

Novel Methods for Elucidating Modality Importance in Multimodal Electrophysiology Classifiers

Multimodal classification is increasingly common in biomedical informatics studies. Many such studies use deep learning classifiers with raw data, which makes explainability difficult. As such, only a few studies have applied explainability methods, and new methods are needed. In this study, we propose sleep stage classification as a testbed for method development and train a convolutional neural network with electroencephalogram (EEG), electrooculogram, and electromyogram data. We then present a global approach that is uniquely adapted for electrophysiology analysis. We further present two local approaches that can identify subject-level differences in explanations that would be obscured by global methods and that can provide insight into the effects of clinical and demographic variables upon the patterns learned by the classifier. We find that EEG is globally the most important modality for all sleep stages, except non-rapid eye movement stage 1 and that local subject-level differences in importance arise. We further show that sex, followed by medication and age had significant effects upon the patterns learned by the classifier. Our novel methods enhance explainability for the growing field of multimodal classification, provide avenues for the advancement of personalized medicine, and yield novel insights into the effects of demographic and clinical variables upon classifiers.

bioinformatics↗

bioRxiv2021

A Model Visualization-based Approach for Insight into Waveforms and Spectra Learned by CNNs

Recent years have shown a growth in the application of deep learning architectures such as convolutional neural networks (CNNs), to electrophysiology analysis. However, using neural networks with raw time-series data makes explainability a significant challenge. Multiple explainability approaches have been developed for insight into the spectral features learned by CNNs from EEG. However, across electrophysiology modalities, and even within EEG, there are many unique waveforms of clinical relevance. Existing methods that provide insight into waveforms learned by CNNs are of questionable utility. In this study, we present a novel model visualization-based approach that analyzes the filters in the first convolutional layer of the network. To our knowledge, this is the first method focused on extracting explainable information from EEG waveforms learned by CNNs while also providing insight into the learned spectral features. We demonstrate the viability of our approach within the context of automated sleep stage classification, a well-characterized domain that can help validate our approach. We identify 3 subgroups of filters with distinct spectral properties, determine the relative importance of each group of filters, and identify several unique waveforms learned by the classifier that were vital to the classifier performance. Our approach represents a significant step forward in explainability for electrophysiology classifiers, which we also hope will be useful for providing insights in future studies. Clinical RelevanceOur approach can assist with the development and validation of clinical time-series classifiers.

bioinformatics↗

bioRxiv2021

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

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.

bioinformatics↗

bioRxiv2021

A Gradient-based Spectral Explainability Method for EEG Deep Learning Classifiers

The automated feature extraction capabilities of deep learning classifiers have promoted their broader application to EEG analysis. In contrast to earlier machine learning studies that used extracted features and traditional explainability approaches, explainability for classifiers trained on raw data is particularly challenging. As such, studies have begun to present methods that provide insight into the spectral features learned by deep learning classifiers trained on raw EEG. These approaches have two key shortcomings. (1) They involve perturbation, which can create out-of-distribution samples that cause inaccurate explanations. (2) They are global, not local. Local explainability approaches can be used to examine how demographic and clinical variables affected the patterns learned by the classifier. In our study, we present a novel local spectral explainability approach. We apply it to a convolutional neural network trained for automated sleep stage classification. We apply layer-wise relevance propagation to identify the relative importance of the features in the raw EEG and subsequently examine the frequency domain of the explanations to determine the importance of each canonical frequency band locally and globally. We then perform a statistical analysis to determine whether age and sex affected the patterns learned by the classifier for each frequency band and sleep stage. Results showed that {delta}, {beta}, and {gamma} were the overall most important frequency bands. In addition, age and sex significantly affected the patterns learned by the classifier for most sleep stages and frequency bands. Our study presents a novel spectral explainability approach that could substantially increase the level of insight into classifiers trained on raw EEG.

bioinformatics↗

bioRxiv2021

A Novel Local Explainability Approach for Spectral Insight into Raw EEG-Based Deep Learning Classifiers

The frequency domain of electroencephalography (EEG) data has developed as a particularly important area of EEG analysis. EEG spectra have been analyzed with explainable machine learning and deep learning methods. However, as deep learning has developed, most studies use raw EEG data, which is not well-suited for traditional explainability methods. Several studies have introduced methods for spectral insight into classifiers trained on raw EEG data. These studies have provided global insight into the frequency bands that are generally important to a classifier but do not provide local insight into the frequency bands important for the classification of individual samples. This local explainability could be particularly helpful for EEG analysis domains like sleep stage classification that feature multiple evolving states. We present a novel local spectral explainability approach and use it to explain a convolutional neural network trained for automated sleep stage classification. We use our approach to show how the relative importance of different frequency bands varies over time and even within the same sleep stages. Furthermore, to better understand how our approach compares to existing methods, we compare a global estimate of spectral importance generated from our local results with an existing global spectral importance approach. We find that the {delta} band is most important for most sleep stages, though {beta} is most important for the non-rapid eye movement 2 (NREM2) sleep stage. Additionally, {theta} is particularly important for identifying Awake and NREM1 samples. Our study represents the first approach developed for local spectral insight into deep learning classifiers trained on raw EEG time series.

bioinformatics↗

bioRxiv2021

A Novel Local Ablation Approach for Explaining Multimodal Classifiers

With the growing use of multimodal data for deep learning classification in healthcare research, more studies have begun to present explainability methods for insight into multimodal classifiers. Among these studies, few have utilized local explainability methods, which could provide (1) insight into the classification of each sample and (2) an opportunity to better understand the effects of demographic and clinical variables within datasets (e.g., medication of subjects in electrophysiology data). To the best of our knowledge, this opportunity has not yet been explored within multimodal classification. We present a novel local ablation approach that shows the importance of each modality to the correct classification of each class and explore the effects of demographic and clinical variables upon the classifier. As a use-case, we train a convolutional neural network for automated sleep staging with electroencephalogram (EEG), electrooculogram (EOG), and electromyogram (EMG) data. We find that EEG is the most important modality across most stages, though EOG is particular important for non-rapid eye movement stage 1. Further, we identify significant relationships between the local explanations and subject age, sex, and state of medication which suggest that the classifier learned specific features associated with these variables across multiple modalities and correctly classified samples. Our novel explainability approach has implications for many fields involving multimodal classification. Moreover, our examination of the degree to which demographic and clinical variables may affect classifiers could provide direction for future studies in automated biomarker discovery.

bioinformatics↗

bioRxiv2021

An Explainable Deep Learning Approach for Multimodal Electrophysiology Classification

In recent years, more biomedical studies have begun to use multimodal data to improve model performance. As such, there is a need for improved multimodal explainability methods. Many studies involving multimodal explainability have used ablation approaches. Ablation requires the modification of input data, which may create out-of-distribution samples and may not always offer a correct explanation. We propose using an alternative gradient-based feature attribution approach, called layer-wise relevance propagation (LRP), to help explain multimodal models. To demonstrate the feasibility of the approach, we selected automated sleep stage classification as our use-case and trained a 1-D convolutional neural network (CNN) with electroencephalogram (EEG), electrooculogram (EOG), and electromyogram (EMG) data. We applied LRP to explain the relative importance of each modality to the classification of different sleep stages. Our results showed that across all samples, EEG was most important, followed by EOG, and EMG. For individual sleep stages, EEG and EOG had higher relevance for classifying awake and non-rapid eye movement 1 (NREM1). EOG was most important for classifying REM, and EEG was most relevant for classifying NREM2-NREM3. Also, LRP gave consistent levels of importance to each modality for correctly classified samples across folds and inconsistent levels of importance for incorrectly classified samples. Our results demonstrate the additional insight that gradient-based approaches can provide relative to ablation methods and highlight their feasibility for explaining multimodal electrophysiology classifiers.

bioinformatics↗

bioRxiv2021

Explainable Sleep Stage Classification with Multimodal Electrophysiology Time-series

Many automated sleep staging studies have used deep learning approaches, and a growing number have used multimodal data to improve their classification performance. However, few studies using multimodal data have provided model explainability. Some have used traditional ablation approaches that "zero out" a modality. However, the samples that result from this ablation are unlikely to be found in real electroencephalography (EEG) data, which could adversely affect the importance estimates that result. Here, we train a convolutional neural network for sleep stage classification with EEG, electrooculograms (EOG), and electromyograms (EMG) and propose an ablation approach that replaces each modality with values that approximate the line-related noise commonly found in electrophysiology data. The relative importance that we identify for each modality is consistent with sleep staging guidelines, with EEG being important for most sleep stages and EOG being important for Rapid Eye Movement (REM) and nonREM stages. EMG showed low relative importance across classes. A comparison of our approach with a "zero out" ablation approach indicates that while the importance results are consistent for the most part, our method accentuates the importance of modalities to the model for the classification of some stages like REM (p < 0.05). These results suggest that a careful, domain-specific selection of an ablation approach may provide a clearer indicator of modality importance. Further, this study provides guidance for future research on using explainability methods with multimodal electrophysiology data.

bioinformatics↗

bioRxiv2021

The link between static and dynamic brain functional network connectivity and genetic risk of Alzheimer's disease

Apolipoprotein E (APOE) polymorphic alleles are genetic factors associated with Alzheimers disease (AD) risk. Although previous studies have explored the link between AD genetic risk and static functional network connectivity (sFNC), to the best of our knowledge, no previous studies have evaluated the association between dynamic FNC (dFNC) and AD genetic risk. Here, we examined the link between sFNC, dFNC, and AD genetic risk with a reproducible, data-driven approach. We used rs-fMRI, demographic, and APOE data from cognitively normal individuals (N=894) between 42 to 95 years of age (mean = 70 years). We divided individuals into low, moderate, and high-risk groups. Using Pearson correlation, we calculated sFNC across seven brain networks. We also calculated dFNC with a sliding window and Pearson correlation. The dFNC windows were partitioned into three distinct states with k-means clustering. Next, we calculated the amount of time each subject spent in each state, called occupancy rate or OCR. We compared both sFNC and OCR, estimated from dFNC, across individuals with different genetic risk and found that both sFNC and dFNC are related to AD genetic risk. We found that higher AD risk reduces within-visual sensory network (VSN) sFNC and that individuals with higher AD risk spend more time in a state with lower within-VSN dFNC. Additionally, we found that AD genetic risk affects whole-brain sFNC and dFNC in women but not in men. In conclusion, we presented novel insights into the links between sFNC, dFNC, and AD genetic risk.

neuroscience↗

bioRxiv2021

The relationship between dynamic functional network connectivity and spatial orientation in healthy young adults

Spatial orientation is essential to interacting with a physical environment, and better understanding it could contribute to a better understanding of a variety of diseases and disorders that are characterized by deficits in spatial orientation. Many previous studies have focused on the relationship between spatial orientation and individual brain regions, though in recent years studies have begun to examine spatial orientation from a network perspective. This study analyzes dynamic functional network connectivity (dFNC) values extracted from over 800 resting-state fMRI recordings of healthy young adults (age 22-37 years) and applies unsupervised machine learning methods to identify neural brain states that occur across all subjects. We estimated the occupancy rate (OCR) for each subject, which was proportional to the amount of time that they spent in each state, and investigated the link between the OCR and spatial orientation and the state-specific FNC values and spatial orientation controlling for age and sex. Our findings showed that the amount of time subjects spent in a state characterized by increased connectivity within and between visual, auditory, and sensorimotor networks and within the default mode network while at rest corresponded to their performance on tests of spatial orientation. We also found that increased sensorimotor network connectivity in two of the identified states negatively correlated with decreased spatial orientation, further highlighting the relationship between the sensorimotor network and spatial orientation. This study provides insight into how the temporal properties of the functional brain connectivity within and between key brain networks may influence spatial orientation.

neuroscience↗

bioRxiv2021

Aberrant Dynamic Functional Connectivity of Default Mode Network in Schizophrenia and Links to Symptom Severity

BackgroundSchizophrenia affects around 1% of the global population. Functional connectivity extracted from resting-state functional magnetic resonance imaging (rs-fMRI) has previously been used to study schizophrenia and has great potential to provide novel insights into the disorder. Some studies have shown abnormal functional connectivity in the default mode network of individuals with schizophrenia, and more recent studies have shown abnormal dynamic functional connectivity (dFC) in individuals with schizophrenia. However, DMN dFC and the link between abnormal DMN dFC and symptom severity have not been well characterized. MethodResting-state fMRI data from subjects with schizophrenia (SZ) and healthy controls (HC) across two datasets were analyzed independently. We captured seven maximally independent subnodes in the DMN by applying group independent component analysis and estimated dFC between subnode time courses using a sliding window approach. A clustering method separated the dFCs into five reoccurring brain states. A feature selection method modeled the difference between SZs and HCs using the state-specific FC features. Finally, we used the transition probability of a hidden Markov model to characterize the link between symptom severity and dFC in SZ subjects. ResultsWe found decreases in the connectivity of the anterior cingulate cortex (ACC) and increases in the connectivity between the precuneus (PCu) and the posterior cingulate cortex (PCC) (i.e., PCu/PCC) of SZ subjects. In SZ, the transition probability from a state with weaker PCu/PCC and stronger ACC connectivity to a state with stronger PCu/PCC and weaker ACC connectivity increased with symptom severity. ConclusionsTo our knowledge, this was the first study to investigate DMN dFC and its link to schizophrenia symptom severity. We identified reproducible neural states in a data-driven manner and demonstrated that the strength of connectivity within those states differed between SZs and HCs. Additionally, we identified a relationship between SZ symptom severity and the dynamics of DMN functional connectivity. We validated our results across two datasets. These results support the potential of dFC for use as a biomarker of schizophrenia and shed new light upon the relationship between schizophrenia and DMN dynamics.

neuroscience↗