An Approach for Estimating Explanation Uncertainty in fMRI dFNC Classification

Abstract

In recent years, many neuroimaging studies have begun to integrate gradient-based explainability methods to provide insight into key features. However, existing explainability approaches typically generate a point estimate of importance and do not provide insight into the degree of uncertainty associated with explanations. In this study, we present a novel approach for estimating explanation uncertainty for convolutional neural networks (CNN) trained on neuroimaging data. We train a CNN for classification of individuals with schizophrenia (SZs) and controls (HCs) using resting state functional magnetic resonance imaging (rs-fMRI) dynamic functional network connectivity (dFNC) data. We apply Monte Carlo batch normalization (MCBN) and generate an explanation following each iteration using layer-wise relevance propagation (LRP). We then examine whether the resulting distribution of explanations differs between SZs and HCs and examine the relationship between MCBN-based LRP and regular LRP. We find a number of significant differences in LRP relevance for SZs and HCs and find that traditional LRP values frequently diverge from the MCBN relevance distribution. This study provides a novel approach for obtaining insight into the level of uncertainty associated with gradient-based explanations in neuroimaging and represents a significant step towards increasing reliability of explainability methods for a clinical setting.

Identifying Neuropsychiatric Disorder Subtypes and Subtype-dependent Variation in Diagnostic Deep Learning Classifier Performance

“…To determine whether the model relied upon different features for identifying each subtype, we used layer-wise relevance propagation (LRP). LRP has been used in a number of neuroscience time-series analyses [9], [15], [16]. We used the αβ-rule (α=1, β=0) to propagate positive relevance through the network (i.e., relevance that a sample belongs to a specific target class, rather than other classes).…”

Towards Greater Neuroimaging Classification Transparency via the Integration of Explainability Methods and Confidence Estimation Approaches

“…To this end, we performed two analyses. (1) We calculated the percentage of samples for each fold that returned Not-a-Number (NaN) importance values for the regular model and for at least one of the 100 iterations of MCD and MCBN. (2) We also calculated the percentage of MCD and MCBN iterations for each sample that produced NaN values across folds.…”

“…In recent years, studies have increasingly sought to develop automated diagnosis approaches using machine learning and deep learning methods for a variety of neurological and neuropsychiatric disorders like schizophrenia [1]- [3], major depressive disorder [4], [5], Alzheimer's disease [6], [7], and others. This growth can be partially attributed to the limitations of existing clinical diagnostic approaches that are often dependent solely upon symptoms, rather than empirical biological markers, for diagnosis [8]- [10].…”

Examining Reproducibility of EEG Schizophrenia Biomarkers Across Explainable Machine Learning Models

“…Much of the research focused on SZ diagnosis and biomarker discovery has involved electroencephalography (EEG), though efforts have been made with other modalities like magnetoencephalography (MEG) [3] and functional magnetic resonance imaging (fMRI) [4], [5]. EEG offers advantages over other modalities in that it is relatively cheap and easy to collect while also having high temporal resolution [6].…”