Decoding and mapping task states of the human brain via deep learning

A Survey on Explainable Artificial Intelligence (XAI) Techniques for Visualizing Deep Learning Models in Medical Imaging

The combination of medical imaging and deep learning has significantly improved diagnostic and prognostic capabilities in the healthcare domain. Nevertheless, the inherent complexity of deep learning models poses challenges in understanding their decision-making processes. Interpretability and visualization techniques have emerged as crucial tools to unravel the black-box nature of these models, providing insights into their inner workings and enhancing trust in their predictions. This survey paper comprehensively examines various interpretation and visualization techniques applied to deep learning models in medical imaging. The paper reviews methodologies, discusses their applications, and evaluates their effectiveness in enhancing the interpretability, reliability, and clinical relevance of deep learning models in medical image analysis.

Explainable deep-learning framework: decoding brain states and prediction of individual performance in false-belief task at early childhood stage

Decoding of cognitive states aims to identify individuals' brain states and brain fingerprints to predict behavior. Deep learning provides an important platform for analyzing brain signals at different developmental stages to understand brain dynamics. Due to their internal architecture and feature extraction techniques, existing machine-learning and deep-learning approaches are suffering from low classification performance and explainability issues that must be improved. In the current study, we hypothesized that even at the early childhood stage (as early as 3-years), connectivity between brain regions could decode brain states and predict behavioral performance in false-belief tasks. To this end, we proposed an explainable deep learning framework to decode brain states (Theory of Mind and Pain states) and predict individual performance on ToM-related false-belief tasks in a developmental dataset. We proposed an explainable spatiotemporal connectivity-based Graph Convolutional Neural Network (Ex-stGCNN) model for decoding brain states. Here, we consider a developmental dataset, N = 155 (122 children; 3-12 yrs and 33 adults; 18-39 yrs), in which participants watched a short, soundless animated movie, shown to activate Theory-of-Mind (ToM) and pain networs. After scanning, the participants underwent a ToM-related false-belief task, leading to categorization into the pass, fail, and inconsistent groups based on performance. We trained our proposed model using Functional Connectivity (FC) and Inter-Subject Functional Correlations (ISFC) matrices separately. We observed that the stimulus-driven feature set (ISFC) could capture ToM and Pain brain states more accurately with an average accuracy of 94%, whereas it achieved 85% accuracy using FC matrices. We also validated our results using five-fold cross-validation and achieved an average accuracy of 92%. Besides this study, we applied the SHapley Additive exPlanations (SHAP) approach to identify brain fingerprints that contributed the most to predictions. We hypothesized that ToM network brain connectivity could predict individual performance on false-belief tasks. We proposed an Explainable Convolutional Variational Auto-Encoder (Ex-Convolutional VAE) model to predict individual performance on false-belief tasks and trained the model using FC and ISFC matrices separately. ISFC matrices again outperformed the FC matrices in prediction of individual performance. We achieved 93.5% accuracy with an F1-score of 0.94 using ISFC matrices and achieved 90% accuracy with an F1-score of 0.91 using FC matrices.

Task sub-type states decoding via group deep bidirectional recurrent neural network

Decoding brain states under different cognitive tasks from functional magnetic resonance imaging (fMRI) data has attracted great attention in the neuroimaging filed. However, the well-known temporal dependency in fMRI sequences has not been fully exploited in existing studies, due to the limited temporal-modeling capacity of the backbone machine learning algorithms and rigid training sample organization strategies upon which the brain decoding methods are built. To address these limitations, we propose a novel method for fine-grain brain state decoding, namely, group deep bidirectional recurrent neural network (Group-DBRNN) model. We first propose a training sample organization strategy that consists of a group-task sample generation module and a multiple-scale random fragment strategy (MRFS) module to collect training samples that contain rich task-relevant brain activity contrast (i.e., the comparison of neural activity patterns between different tasks) and maintain the temporal dependency. We then develop a novel decoding model by replacing the unidirectional RNNs that are widely used in existing brain state decoding studies with bidirectional stacked RNNs to better capture the temporal dependency, and by introducing a multi-task interaction layer (MTIL) module to effectively model the task-relevant brain activity contrast. Our experimental results on the Human Connectome Project task fMRI dataset (7 tasks consisting of 23 task sub-type states) show that the proposed model achieves an average decoding accuracy of 94.7% over the 23 fine-grain sub-type states. Meanwhile, our extensive interpretations of the intermediate features learned in the proposed model via visualizations and quantitative assessments of their discriminability and inter-subject alignment evidence that the proposed model can effectively capture the temporal dependency and task-relevant contrast.

Unsupervised Joint Domain Adaptation for Decoding Brain Cognitive States From tfMRI Images

Recent advances in large model and neuroscience have enabled exploration of the mechanism of brain activity by using neuroimaging data. Brain decoding is one of the most promising researches to further understand the human cognitive function. However, current methods excessively depends on high-quality labeled data, which brings enormous expense of collection and annotation of neural images by experts. Besides, the performance of cross-individual decoding suffers from inconsistency in data distribution caused by individual variation and different collection equipments. To address mentioned above issues, a Join Domain Adapative Decoding (JDAD) framework is proposed for unsupervised decoding specific brain cognitive state related to behavioral task. Based on the volumetric feature extraction from task-based functional Magnetic Resonance Imaging (tfMRI) data, a novel objective loss function is designed by the combination of joint distribution regularizer, which aims to restrict the distance of both the conditional and marginal probability distribution of labeled and unlabeled samples. Experimental results on the public Human Connectome Project (HCP) S1200 dataset show that JDAD achieves superior performance than other prevalent methods, especially for fine-grained task with 11.5%-21.6% improvements of decoding accuracy. The learned 3D features are visualized by Grad-CAM to build a combination with brain functional regions, which provides a novel path to learn the function of brain cortex regions related to specific cognitive task in group level.

Brain-like illusion produced by Skye's Oblique Grating in deep neural networks

The analogy between the brain and deep neural networks (DNNs) has sparked interest in neuroscience. Although DNNs have limitations, they remain valuable for modeling specific brain characteristics. This study used Skye's Oblique Grating illusion to assess DNNs' relevance to brain neural networks. We collected data on human perceptual responses to a series of visual illusions. This data was then used to assess how DNN responses to these illusions paralleled or differed from human behavior. We performed two analyses:(1) We trained DNNs to perform horizontal vs. non-horizontal classification on images with bars tilted different degrees (non-illusory images) and tested them on images with horizontal bars with different illusory strengths measured by human behavior (illusory images), finding that DNNs showed human-like illusions; (2) We performed representational similarity analysis to assess whether illusory representation existed in different layers within DNNs, finding that DNNs showed illusion-like responses to illusory images. The representational similarity between real tilted images and illusory images was calculated, which showed the highest values in the early layers and decreased layer-by-layer. Our findings suggest that DNNs could serve as potential models for explaining the mechanism of visual illusions in human brain, particularly those that may originate in early visual areas like the primary visual cortex (V1). While promising, further research is necessary to understand the nuanced differences between DNNs and human visual pathways.

A hybrid spatiotemporal deep belief network and sparse representation-based framework reveals multilevel core functional components in decoding multitask fMRI signals

Decoding human brain activity on various task-based functional brain imaging data is of great significance for uncovering the functioning mechanism of the human mind. Currently, most feature extraction model-based methods for brain state decoding are shallow machine learning models, which may struggle to capture complex and precise spatiotemporal patterns of brain activity from the highly noisy fMRI raw data. Moreover, although decoding models based on deep learning methods benefit from their multilayer structure that could extract spatiotemporal features at multiscale, the relatively large populations of fMRI datasets are indispensable, and the explainability of their results is elusive. To address the above problems, we proposed a computational framework based on hybrid spatiotemporal deep belief network and sparse representations to differentiate multitask fMRI (tfMRI) signals. Using a relatively small cohort of tfMRI data as a test bed, our framework can achieve an average classification accuracy of 97.86% and define the multilevel temporal and spatial patterns of multiple cognitive tasks. Intriguingly, our model can characterize the key components for differentiating the multitask fMRI signals. Overall, the proposed framework can identify the interpretable and discriminative fMRI composition patterns at multiple scales, offering an effective methodology for basic neuroscience and clinical research with relatively small cohorts.

High-accuracy machine learning techniques for functional connectome fingerprinting and cognitive state decoding

The human brain is a complex network comprised of functionally and anatomically interconnected brain regions. A growing number of studies have suggested that empirical estimates of brain networks may be useful for discovery of biomarkers of disease and cognitive state. A prerequisite for realizing this aim, however, is that brain networks also serve as reliable markers of an individual. Here, using Human Connectome Project data, we build upon recent studies examining brain-based fingerprints of individual subjects and cognitive states based on cognitively demanding tasks that assess, for example, working memory, theory of mind, and motor function. Our approach achieves accuracy of up to 99% for both identification of the subject of an fMRI scan, and for classification of the cognitive state of a previously unseen subject in a scan. More broadly, we explore the accuracy and reliability of five different machine learning techniques on subject fingerprinting and cognitive state decoding objectives, using functional connectivity data from fMRI scans of a high number of subjects (865) across a number of cognitive states (8). These results represent an advance on existing techniques for functional connectivity-based brain fingerprinting and state decoding. Additionally, 16 different functional connectome (FC) matrix construction pipelines are compared in order to characterize the effects of different aspects of the production of FCs on the accuracy of subject and task classification, and to identify possible confounds.

A Deep Learning Model for Correlation Analysis between Electroencephalography Signal and Speech Stimuli

In recent years, the use of electroencephalography (EEG) has grown as a tool for diagnostic and brain function monitoring, being a simple and non-invasive method compared with other procedures like histological sampling. Typically, in order to extract functional brain responses from EEG signals, prolonged and repeated stimuli are needed because of the artifacts generated in recordings which adversely impact the stimulus-response analysis. To mitigate the artifact effect, correlation analysis (CA) methods are applied in the literature, where the predominant approaches focus on enhancing stimulus-response correlations through the use of linear analysis methods like canonical correlation analysis (CCA). This paper introduces a novel CA framework based on a neural network with a loss function specifically designed to maximize correlation between EEG and speech stimuli. Compared with other deep learning CA approaches (DCCAs) in the literature, this framework introduces a single multilayer perceptron (MLP) network instead of two networks for each stimulus. To validate the proposed approach, a comparison with linear CCA (LCCA) and DCCA was performed, using a dataset containing the EEG traces of subjects listening to speech stimuli. The experimental results show that the proposed method improves the overall Pearson correlation by 10.56% compared with the state-of-the-art DCCA method.

ConnSearch: A framework for functional connectivity analysis designed for interpretability and effectiveness at limited sample sizes

Functional connectivity studies increasingly turn to machine learning methods, which typically involve fitting a connectome-wide classifier, then conducting post hoc interpretation analyses to identify the neural correlates that best predict a dependent variable. However, this traditional analytic paradigm suffers from two main limitations. First, even if classifiers are perfectly accurate, interpretation analyses may not identify all the patterns expressed by a dependent variable. Second, even if classifiers are generalizable, the patterns implicated via interpretation analyses may not replicate. In other words, this traditional approach can yield effective classifiers while falling short of most neuroscientists' goals: pinpointing the neural correlates of dependent variables. We propose a new framework for multivariate analysis, ConnSearch, which involves dividing the connectome into components (e.g., groups of highly connected regions) and fitting an independent model for each component (e.g., a support vector machine or a correlation-based model). Conclusions about the link between a dependent variable and the brain are based on which components yield predictive models rather than on interpretation analysis. We used working memory data from the Human Connectome Project (N = 50-250) to compare ConnSearch with four existing connectome-wide classification/interpretation methods. For each approach, the models attempted to classify examples as being from the high-load or low-load conditions (binary labels). Relative to traditional methods, ConnSearch identified neural correlates that were more comprehensive, had greater consistency with the WM literature, and better replicated across datasets. Hence, ConnSearch is well-positioned to be an effective tool for functional connectivity research.

Self-supervised pretraining improves the performance of classification of task functional magnetic resonance imaging

Introduction

Decoding brain activities is one of the most popular topics in neuroscience in recent years. And deep learning has shown high performance in fMRI data classification and regression, but its requirement for large amounts of data conflicts with the high cost of acquiring fMRI data.

Methods

In this study, we propose an end-to-end temporal contrastive self-supervised learning algorithm, which learns internal spatiotemporal patterns within fMRI and allows the model to transfer learning to datasets of small size. For a given fMRI signal, we segmented it into three sections: the beginning, middle, and end. We then utilized contrastive learning by taking the end-middle (i.e., neighboring) pair as the positive pair, and the beginning-end (i.e., distant) pair as the negative pair.

Results

We pretrained the model on 5 out of 7 tasks from the Human Connectome Project (HCP) and applied it in a downstream classification of the remaining two tasks. The pretrained model converged on data from 12 subjects, while a randomly initialized model required 100 subjects. We then transferred the pretrained model to a dataset containing unpreprocessed whole-brain fMRI from 30 participants, achieving an accuracy of 80.2 ± 4.7%, while the randomly initialized model failed to converge. We further validated the model's performance on the Multiple Domain Task Dataset (MDTB), which contains fMRI data of 26 tasks from 24 participants. Thirteen tasks of fMRI were selected as inputs, and the results showed that the pre-trained model succeeded in classifying 11 of the 13 tasks. When using the 7 brain networks as input, variations of the performance were observed, with the visual network performed as well as whole brain inputs, while the limbic network almost failed in all 13 tasks.

Discussion

Our results demonstrated the potential of self-supervised learning for fMRI analysis with small datasets and unpreprocessed data, and for analysis of the correlation between regional fMRI activity and cognitive tasks.

Benchmarking explanation methods for mental state decoding with deep learning models

Deep learning (DL) models find increasing application in mental state decoding, where researchers seek to understand the mapping between mental states (e.g., experiencing anger or joy) and brain activity by identifying those spatial and temporal features of brain activity that allow to accurately identify (i.e., decode) these states. Once a DL model has been trained to accurately decode a set of mental states, neuroimaging researchers often make use of methods from explainable artificial intelligence research to understand the model's learned mappings between mental states and brain activity. Here, we benchmark prominent explanation methods in a mental state decoding analysis of multiple functional Magnetic Resonance Imaging (fMRI) datasets. Our findings demonstrate a gradient between two key characteristics of an explanation in mental state decoding, namely, its faithfulness and its alignment with other empirical evidence on the mapping between brain activity and decoded mental state: explanation methods with high explanation faithfulness, which capture the model's decision process well, generally provide explanations that align less well with other empirical evidence than the explanations of methods with less faithfulness. Based on our findings, we provide guidance for neuroimaging researchers on how to choose an explanation method to gain insight into the mental state decoding decisions of DL models.

Multimodal deep neural decoding reveals highly resolved spatiotemporal profile of visual object representation in humans

Perception and categorization of objects in a visual scene are essential to grasp the surrounding situation. Recently, neural decoding schemes, such as machine learning in functional magnetic resonance imaging (fMRI), has been employed to elucidate the underlying neural mechanisms. However, it remains unclear as to how spatially distributed brain regions temporally represent visual object categories and sub-categories. One promising strategy to address this issue is neural decoding with concurrently obtained neural response data of high spatial and temporal resolution. In this study, we explored the spatial and temporal organization of visual object representations using concurrent fMRI and electroencephalography (EEG), combined with neural decoding using deep neural networks (DNNs). We hypothesized that neural decoding by multimodal neural data with DNN would show high classification performance in visual object categorization (faces or non-face objects) and sub-categorization within faces and objects. Visualization of the fMRI DNN was more sensitive than that in the univariate approach and revealed that visual categorization occurred in brain-wide regions. Interestingly, the EEG DNN valued the earlier phase of neural responses for categorization and the later phase of neural responses for sub-categorization. Combination of the two DNNs improved the classification performance for both categorization and sub-categorization compared with fMRI DNN or EEG DNN alone. These deep learning-based results demonstrate a categorization principle in which visual objects are represented in a spatially organized and coarse-to-fine manner, and provide strong evidence of the ability of multimodal deep learning to uncover spatiotemporal neural machinery in sensory processing.

Recent Advances in Explainable Artificial Intelligence for Magnetic Resonance Imaging

Advances in artificial intelligence (AI), especially deep learning (DL), have facilitated magnetic resonance imaging (MRI) data analysis, enabling AI-assisted medical image diagnoses and prognoses. However, most of the DL models are considered as "black boxes". There is an unmet need to demystify DL models so domain experts can trust these high-performance DL models. This has resulted in a sub-domain of AI research called explainable artificial intelligence (XAI). In the last decade, many experts have dedicated their efforts to developing novel XAI methods that are competent at visualizing and explaining the logic behind data-driven DL models. However, XAI techniques are still in their infancy for medical MRI image analysis. This study aims to outline the XAI applications that are able to interpret DL models for MRI data analysis. We first introduce several common MRI data modalities. Then, a brief history of DL models is discussed. Next, we highlight XAI frameworks and elaborate on the principles of multiple popular XAI methods. Moreover, studies on XAI applications in MRI image analysis are reviewed across the tissues/organs of the human body. A quantitative analysis is conducted to reveal the insights of MRI researchers on these XAI techniques. Finally, evaluations of XAI methods are discussed. This survey presents recent advances in the XAI domain for explaining the DL models that have been utilized in MRI applications.

Framework for Accurate Classification of Self-Reported Stress From Multisession Functional MRI Data of Veterans With Posttraumatic Stress

Background: Posttraumatic stress disorder (PTSD) is a significant burden among combat Veterans returning from the wars in Iraq and Afghanistan. While empirically supported treatments have demonstrated reductions in PTSD symptomatology, there remains a need to improve treatment effectiveness. Functional magnetic resonance imaging (fMRI) neurofeedback has emerged as a possible treatment to ameliorate PTSD symptom severity. Virtual reality (VR) approaches have also shown promise in increasing treatment compliance and outcomes. To facilitate fMRI neurofeedback-associated therapies, it would be advantageous to accurately classify internal brain stress levels while Veterans are exposed to trauma-associated VR imagery. Methods: Across 2 sessions, we used fMRI to collect neural responses to trauma-associated VR-like stimuli among male combat Veterans with PTSD symptoms (N = 8). Veterans reported their self-perceived stress level on a scale from 1 to 8 every 15 s throughout the fMRI sessions. In our proposed framework, we precisely sample the fMRI data on cortical gray matter, blurring the data along the gray-matter manifold to reduce noise and dimensionality while preserving maximum neural information. Then, we independently applied 3 machine learning (ML) algorithms to this fMRI data collected across 2 sessions, separately for each Veteran, to build individualized ML models that predicted their internal brain states (self-reported stress responses). Results: We accurately classified the 8-class self-reported stress responses with a mean (± standard error) root mean square error of 0.6 (± 0.1) across all Veterans using the best ML approach. Conclusions: The findings demonstrate the predictive ability of ML algorithms applied to whole-brain cortical fMRI data collected during individual Veteran sessions. The framework we have developed to preprocess whole-brain cortical fMRI data and train ML models across sessions would provide a valuable tool to enable individualized real-time fMRI neurofeedback during VR-like exposure therapy for PTSD.

An explainable deep learning framework for characterizing and interpreting human brain states

Deep learning approaches have been widely adopted in the medical image analysis field. However, a most of existing deep learning approaches focus on achieving promising performances such as classification, detection, and segmentation, and much less effort is devoted to the explanation of the designed models. Similarly, in the brain imaging field, many deep learning approaches have been designed and applied to characterize and predict human brain states. However, these models lack interpretation. In response, we propose a novel domain knowledge informed self-attention graph pooling-based (SAGPool) graph convolutional neural network to study human brain states. Specifically, the dense individualized and common connectivity-based cortical landmarks system (DICCCOL, structural brain connectivity profiles) and holistic atlases of functional networks and interactions system (HAFNI, functional brain connectivity profiles) are integrated with the SAGPool model to better characterize and interpret the brain states. Extensive experiments are designed and carried out on the large-scale human connectome project (HCP) Q1 and S1200 dataset. Promising brain state classification performances are observed (e.g., an average of 93.7% for seven-task classification and 100% for binary classification). In addition, the importance of the brain regions, which contributes most to the accurate classification, is successfully quantified and visualized. A thorough neuroscientific interpretation suggests that these extracted brain regions and their importance calculated from self-attention graph pooling layer offer substantial explainability.

On the benefits of self-taught learning for brain decoding

CONTEXT

We study the benefits of using a large public neuroimaging database composed of functional magnetic resonance imaging (fMRI) statistic maps, in a self-taught learning framework, for improving brain decoding on new tasks. First, we leverage the NeuroVault database to train, on a selection of relevant statistic maps, a convolutional autoencoder to reconstruct these maps. Then, we use this trained encoder to initialize a supervised convolutional neural network to classify tasks or cognitive processes of unseen statistic maps from large collections of the NeuroVault database.

RESULTS

We show that such a self-taught learning process always improves the performance of the classifiers, but the magnitude of the benefits strongly depends on the number of samples available both for pretraining and fine-tuning the models and on the complexity of the targeted downstream task.

CONCLUSION

The pretrained model improves the classification performance and displays more generalizable features, less sensitive to individual differences.

Interpreting mental state decoding with deep learning models

In mental state decoding, researchers aim to identify the set of mental states (e.g., experiencing happiness or fear) that can be reliably identified from the activity patterns of a brain region (or network). Deep learning (DL) models are highly promising for mental state decoding because of their unmatched ability to learn versatile representations of complex data. However, their widespread application in mental state decoding is hindered by their lack of interpretability, difficulties in applying them to small datasets, and in ensuring their reproducibility and robustness. We recommend approaching these challenges by leveraging recent advances in explainable artificial intelligence (XAI) and transfer learning, and also provide recommendations on how to improve the reproducibility and robustness of DL models in mental state decoding.

Deep learning in neuroimaging data analysis: Applications, challenges, and solutions

Methods for the analysis of neuroimaging data have advanced significantly since the beginning of neuroscience as a scientific discipline. Today, sophisticated statistical procedures allow us to examine complex multivariate patterns, however most of them are still constrained by assuming inherent linearity of neural processes. Here, we discuss a group of machine learning methods, called deep learning, which have drawn much attention in and outside the field of neuroscience in recent years and hold the potential to surpass the mentioned limitations. Firstly, we describe and explain the essential concepts in deep learning: the structure and the computational operations that allow deep models to learn. After that, we move to the most common applications of deep learning in neuroimaging data analysis: prediction of outcome, interpretation of internal representations, generation of synthetic data and segmentation. In the next section we present issues that deep learning poses, which concerns multidimensionality and multimodality of data, overfitting and computational cost, and propose possible solutions. Lastly, we discuss the current reach of DL usage in all the common applications in neuroimaging data analysis, where we consider the promise of multimodality, capability of processing raw data, and advanced visualization strategies. We identify research gaps, such as focusing on a limited number of criterion variables and the lack of a well-defined strategy for choosing architecture and hyperparameters. Furthermore, we talk about the possibility of conducting research with constructs that have been ignored so far or/and moving toward frameworks, such as RDoC, the potential of transfer learning and generation of synthetic data.

Decoding Task-Based fMRI Data with Graph Neural Networks, Considering Individual Differences

Task fMRI provides an opportunity to analyze the working mechanisms of the human brain during specific experimental paradigms. Deep learning models have increasingly been applied for decoding and encoding purposes study to representations in task fMRI data. More recently, graph neural networks, or neural networks models designed to leverage the properties of graph representations, have recently shown promise in task fMRI decoding studies. Here, we propose an end-to-end graph convolutional network (GCN) framework with three convolutional layers to classify task fMRI data from the Human Connectome Project dataset. We compared the predictive performance of our GCN model across four of the most widely used node embedding algorithms-NetMF, RandNE, Node2Vec, and Walklets-to automatically extract the structural properties of the nodes in the functional graph. The empirical results indicated that our GCN framework accurately predicted individual differences (0.978 and 0.976) with the NetMF and RandNE embedding methods, respectively. Furthermore, to assess the effects of individual differences, we tested the classification performance of the model on sub-datasets divided according to gender and fluid intelligence. Experimental results indicated significant differences in the classification predictions of gender, but not high/low fluid intelligence fMRI data. Our experiments yielded promising results and demonstrated the superior ability of our GCN in modeling task fMRI data.

A semi-supervised classification RBM with an improved fMRI representation algorithm

BACKGROUND AND OBJECTIVE

Training an effective and robust supervised learning classifier is not easy due to the limitations of acquiring and labeling considerable human functional magnetic resonance imaging (fMRI) data. Semi-supervised learning uses unlabeled data for feature learning and combines them into labeled data to build better classification models.

METHODS

Since no premises or assumptions are required, a restricted Boltzmann machine (RBM) is suitable for learning data representation of neuroimages. In our study, an improved fMRI representation algorithm with a hybrid L1/L2 regularization method (HRBM) was proposed to optimize the original model for sparsity. Different from common semi-supervised classification models that treat feature learning and classification as two separate training steps, we then constructed a new semi-supervised classification RBM based on a joint training algorithm with HRBM, named Semi-HRBM. This joint training algorithm jointly trains the objective function of feature learning and classification process, so that the learned features can effectively represent the original fMRI data and adapt to the classification tasks.

RESULTS

This study uses fMRI data to identify categories of visual stimuli. In the fMRI data classification task under four visual stimuli (house, face, car, and cat), our HRBM has satisfactory feature representation capabilities and better performance for multiple classification tasks. Taking the supervised RBM (sup-RBM) as an example, our Semi-HRBM classification model improves the average accuracy of the four-classification task by 7.68%, and improves the average F1 score of each visual stimulus task by 8.90%. In addition, the generalization ability of the model was also improved.

CONCLUSION

This research might contribute to enrich solutions for insufficiently labeled neuroimaging samples, which could help to identify complex brain states under different stimuli or tasks.

Toward Precise Localization of Abnormal Brain Activity: 1D CNN on Single Voxel fMRI Time-Series

Functional magnetic resonance imaging (fMRI) is one of the best techniques for precise localization of abnormal brain activity non-invasively. Machine-learning approaches have been widely used in neuroimaging studies; however, few studies have investigated the single-voxel modeling of fMRI data under cognitive tasks. We proposed a hybrid one-dimensional (1D) convolutional neural network (1D-CNN) based on the temporal dynamics of single-voxel fMRI time-series and successfully differentiated two continuous task states, namely, self-initiated (SI) and visually guided (VG) motor tasks. First, 25 activation peaks were identified from the contrast maps of SI and VG tasks in a blocked design. Then, the fMRI time-series of each peak voxel was transformed into a temporal-frequency domain by using continuous wavelet transform across a broader frequency range (0.003–0.313 Hz, with a step of 0.01 Hz). The transformed time-series was inputted into a 1D-CNN model for the binary classification of SI and VG continuous tasks. Compared with the univariate analysis, e.g., amplitude of low-frequency fluctuation (ALFF) at each frequency band, including, wavelet-ALFF, the 1D-CNN model highly outperformed wavelet-ALFF, with more efficient decoding models [46% of 800 models showing area under the curve (AUC) > 0.61] and higher decoding accuracies (94% of the efficient models), especially on the high-frequency bands (>0.1 Hz). Moreover, our results also demonstrated the advantages of wavelet decompositions over the original fMRI series by showing higher decoding performance on all peak voxels. Overall, this study suggests a great potential of single-voxel analysis using 1D-CNN and wavelet transformation of fMRI series with continuous, naturalistic, steady-state task design or resting-state design. It opens new avenues to precise localization of abnormal brain activity and fMRI-guided precision brain stimulation therapy.

Counterfactual Explanation of Brain Activity Classifiers Using Image-To-Image Transfer by Generative Adversarial Network

Deep neural networks (DNNs) can accurately decode task-related information from brain activations. However, because of the non-linearity of DNNs, it is generally difficult to explain how and why they assign certain behavioral tasks to given brain activations, either correctly or incorrectly. One of the promising approaches for explaining such a black-box system is counterfactual explanation. In this framework, the behavior of a black-box system is explained by comparing real data and realistic synthetic data that are specifically generated such that the black-box system outputs an unreal outcome. The explanation of the system's decision can be explained by directly comparing the real and synthetic data. Recently, by taking advantage of advances in DNN-based image-to-image translation, several studies successfully applied counterfactual explanation to image domains. In principle, the same approach could be used in functional magnetic resonance imaging (fMRI) data. Because fMRI datasets often contain multiple classes (e.g., multiple behavioral tasks), the image-to-image transformation applicable to counterfactual explanation needs to learn mapping among multiple classes simultaneously. Recently, a new generative neural network (StarGAN) that enables image-to-image transformation among multiple classes has been developed. By adapting StarGAN with some modifications, here, we introduce a novel generative DNN (counterfactual activation generator, CAG) that can provide counterfactual explanations for DNN-based classifiers of brain activations. Importantly, CAG can simultaneously handle image transformation among all the seven classes in a publicly available fMRI dataset. Thus, CAG could provide a counterfactual explanation of DNN-based multiclass classifiers of brain activations. Furthermore, iterative applications of CAG were able to enhance and extract subtle spatial brain activity patterns that affected the classifier's decisions. Together, these results demonstrate that the counterfactual explanation based on image-to-image transformation would be a promising approach to understand and extend the current application of DNNs in fMRI analyses.

Brain Decoding Using fMRI Images for Multiple Subjects through Deep Learning

Substantial information related to human cerebral conditions can be decoded through various noninvasive evaluating techniques like fMRI. Exploration of the neuronal activity of the human brain can divulge the thoughts of a person like what the subject is perceiving, thinking, or visualizing. Furthermore, deep learning techniques can be used to decode the multifaceted patterns of the brain in response to external stimuli. Existing techniques are capable of exploring and classifying the thoughts of the human subject acquired by the fMRI imaging data. fMRI images are the volumetric imaging scans which are highly dimensional as well as require a lot of time for training when fed as an input in the deep learning network. However, the hassle for more efficient learning of highly dimensional high-level features in less training time and accurate interpretation of the brain voxels with less misclassification error is needed. In this research, we propose an improved CNN technique where features will be functionally aligned. The optimal features will be selected after dimensionality reduction. The highly dimensional feature vector will be transformed into low dimensional space for dimensionality reduction through autoadjusted weights and combination of best activation functions. Furthermore, we solve the problem of increased training time by using Swish activation function, making it denser and increasing efficiency of the model in less training time. Finally, the experimental results are evaluated and compared with other classifiers which demonstrated the supremacy of the proposed model in terms of accuracy.

Attention module improves both performance and interpretability of four-dimensional functional magnetic resonance imaging decoding neural network

Decoding brain cognitive states from neuroimaging signals is an important topic in neuroscience. In recent years, deep neural networks (DNNs) have been recruited for multiple brain state decoding and achieved good performance. However, the open question of how to interpret the DNN black box remains unanswered. Capitalizing on advances in machine learning, we integrated attention modules into brain decoders to facilitate an in‐depth interpretation of DNN channels. A four‐dimensional (4D) convolution operation was also included to extract temporo‐spatial interaction within the fMRI signal. The experiments showed that the proposed model obtains a very high accuracy (97.4%) and outperforms previous researches on the seven different task benchmarks from the Human Connectome Project (HCP) dataset. The visualization analysis further illustrated the hierarchical emergence of task‐specific masks with depth. Finally, the model was retrained to regress individual traits within the HCP and to classify viewing images from the BOLD5000 dataset, respectively. Transfer learning also achieves good performance. Further visualization analysis shows that, after transfer learning, low‐level attention masks remained similar to the source domain, whereas high‐level attention masks changed adaptively. In conclusion, the proposed 4D model with attention module performed well and facilitated interpretation of DNNs, which is helpful for subsequent research.

Representation learning of resting state fMRI with variational autoencoder

Resting state functional magnetic resonance imaging (rsfMRI) data exhibits complex but structured patterns. However, the underlying origins are unclear and entangled in rsfMRI data. Here we establish a variational auto-encoder, as a generative model trainable with unsupervised learning, to disentangle the unknown sources of rsfMRI activity. After being trained with large data from the Human Connectome Project, the model has learned to represent and generate patterns of cortical activity and connectivity using latent variables. The latent representation and its trajectory represent the spatiotemporal characteristics of rsfMRI activity. The latent variables reflect the principal gradients of the latent trajectory and drive activity changes in cortical networks. Representational geometry captured as covariance or correlation between latent variables, rather than cortical connectivity, can be used as a more reliable feature to accurately identify subjects from a large group, even if only a short period of data is available in each subject. Our results demonstrate that VAE is a valuable addition to existing tools, particularly suited for unsupervised representation learning of resting state fMRI activity.

Structure can predict function in the human brain: a graph neural network deep learning model of functional connectivity and centrality based on structural connectivity

Although functional connectivity and associated graph theory measures (e.g., centrality; how centrally important to the network a region is) are widely used in brain research, the full extent to which these functional measures are related to the underlying structural connectivity is not yet fully understood. Graph neural network deep learning methods have not yet been applied for this purpose, and offer an ideal model architecture for working with connectivity data given their ability to capture and maintain inherent network structure. Here, we applied this model to predict functional connectivity from structural connectivity in a sample of 998 participants from the Human Connectome Project. Our results showed that the graph neural network accounted for 89% of the variance in mean functional connectivity, 56% of the variance in individual-level functional connectivity, 99% of the variance in mean functional centrality, and 81% of the variance in individual-level functional centrality. These results represent an important finding that functional centrality can be robustly predicted from structural connectivity. Regions of particular importance to the model's performance as determined through lesioning are discussed, whereby regions with higher centrality have a higher impact on model performance. Future research on models of patient, demographic, or behavioural data can also benefit from this graph neural network method as it is ideally-suited for depicting connectivity and centrality in brain networks. These results have set a new benchmark for prediction of functional connectivity from structural connectivity, and models like this may ultimately lead to a way to predict functional connectivity in individuals who are unable to do fMRI tasks (e.g., non-responsive patients).

Reversible Fronto-occipitotemporal Signaling Complements Task Encoding and Switching under Ambiguous Cues

Adaptation to changing environments involves the appropriate extraction of environmental information to achieve a behavioral goal. It remains unclear how behavioral flexibility is guided under situations where the relevant behavior is ambiguous. Using functional brain mapping of machine learning decoders and directional functional connectivity, we show that brain-wide reversible neural signaling underpins task encoding and behavioral flexibility in ambiguously changing environments. When relevant behavior is cued ambiguously during behavioral shifting, neural coding is attenuated in distributed cortical regions, but top-down signals from the prefrontal cortex complement the coding. When behavioral shifting is cued more explicitly, modality-specialized occipitotemporal regions implement distinct neural coding about relevant behavior, and bottom-up signals from the occipitotemporal region to the prefrontal cortex supplement the behavioral shift. These results suggest that our adaptation to an ever-changing world is orchestrated by the alternation of top-down and bottom-up signaling in the fronto-occipitotemporal circuit depending on the availability of environmental information.

Design of Deep Learning Model for Task-Evoked fMRI Data Classification

Machine learning methods have been successfully applied to neuroimaging signals, one of which is to decode specific task states from functional magnetic resonance imaging (fMRI) data. In this paper, we propose a model that simultaneously utilizes characteristics of both spatial and temporal sequential information of fMRI data with deep neural networks to classify the fMRI task states. We designed a convolution network module and a recurrent network module to extract the spatial and temporal features of fMRI data, respectively. In particular, we also add the attention mechanism to the recurrent network module, which more effectively highlights the brain activation state at the moment of reaction. We evaluated the model using task-evoked fMRI data from the Human Connectome Project (HCP) dataset, the classification accuracy got 94.31%, and the experimental results have shown that the model can effectively distinguish the brain states under different task stimuli.

A State-of-the-Art of Functional Scaffolds for 3D Nervous Tissue Regeneration

Exploring and developing multifunctional intelligent biomaterials is crucial to improve next-generation therapies in tissue engineering and regenerative medicine. Recent findings show how distinct characteristics of in situ microenvironment can be mimicked by using different biomaterials. In vivo tissue architecture is characterized by the interconnection between cells and specific components of the extracellular matrix (ECM). Last evidence shows the importance of the structure and composition of the ECM in the development of cellular and molecular techniques, to achieve the best biodegradable and bioactive biomaterial compatible to human physiology. Such biomaterials provide specialized bioactive signals to regulate the surrounding biological habitat, through the progression of wound healing and biomaterial integration. The connection between stem cells and biomaterials stimulate the occurrence of specific modifications in terms of cell properties and fate, influencing then processes such as self-renewal, cell adhesion and differentiation. Recent studies in the field of tissue engineering and regenerative medicine have shown to deal with a broad area of applications, offering the most efficient and suitable strategies to neural repair and regeneration, drawing attention towards the potential use of biomaterials as 3D tools for in vitro neurodevelopment of tissue models, both in physiological and pathological conditions. In this direction, there are several tools supporting cell regeneration, which associate cytokines and other soluble factors delivery through the scaffold, and different approaches considering the features of the biomaterials, for an increased functionalization of the scaffold and for a better promotion of neural proliferation and cells-ECM interplay. In fact, 3D scaffolds need to ensure a progressive and regular delivery of cytokines, growth factors, or biomolecules, and moreover they should serve as a guide and support for injured tissues. It is also possible to create scaffolds with different layers, each one possessing different physical and biochemical aspects, able to provide at the same time organization, support and maintenance of the specific cell phenotype and diversified ECM morphogenesis. Our review summarizes the most recent advancements in functional materials, which are crucial to achieve the best performance and at the same time, to overcome the current limitations in tissue engineering and nervous tissue regeneration.

Evaluation of Task fMRI Decoding With Deep Learning on a Small Sample Dataset

Recently, several deep learning methods have been applied to decoding in task-related fMRI, and their advantages have been exploited in a variety of ways. However, this paradigm is sometimes problematic, due to the difficulty of applying deep learning to high-dimensional data and small sample size conditions. The difficulties in gathering a large amount of data to develop predictive machine learning models with multiple layers from fMRI experiments with complicated designs and tasks are well-recognized. Group-level, multi-voxel pattern analysis with small sample sizes results in low statistical power and large accuracy evaluation errors; failure in such instances is ascribed to the individual variability that risks information leakage, a particular issue when dealing with a limited number of subjects. In this study, using a small-size fMRI dataset evaluating bilingual language switch in a property generation task, we evaluated the relative fit of different deep learning models, incorporating moderate split methods to control the amount of information leakage. Our results indicated that using the session shuffle split as the data folding method, along with the multichannel 2D convolutional neural network (M2DCNN) classifier, recorded the best authentic classification accuracy, which outperformed the efficiency of 3D convolutional neural network (3DCNN). In this manuscript, we discuss the tolerability of within-subject or within-session information leakage, of which the impact is generally considered small but complex and essentially unknown; this requires clarification in future studies.

Pooling Regularized Graph Neural Network for fMRI Biomarker Analysis

Understanding how certain brain regions relate to a specific neurological disorder has been an important area of neuroimaging research. A promising approach to identify the salient regions is using Graph Neural Networks (GNNs), which can be used to analyze graph structured data, e.g. brain networks constructed by functional magnetic resonance imaging (fMRI). We propose an interpretable GNN framework with a novel salient region selection mechanism to determine neurological brain biomarkers associated with disorders. Specifically, we design novel regularized pooling layers that highlight salient regions of interests (ROIs) so that we can infer which ROIs are important to identify a certain disease based on the node pooling scores calculated by the pooling layers. Our proposed framework, Pooling Regularized-GNN (PR-GNN), encourages reasonable ROI-selection and provides flexibility to preserve either individual- or group-level patterns. We apply the PR-GNN framework on a Biopoint Autism Spectral Disorder (ASD) fMRI dataset. We investigate different choices of the hyperparameters and show that PR-GNN outperforms baseline methods in terms of classification accuracy. The salient ROI detection results show high correspondence with the previous neuroimaging-derived biomarkers for ASD.

Deep learning object detection of maxillary cyst-like lesions on panoramic radiographs: preliminary study

OBJECTIVES

This study aimed to examine the performance of deep learning object detection technology for detecting and identifying maxillary cyst-like lesions on panoramic radiography.

METHODS

Altogether, 412 patients with maxillary cyst-like lesions (including several benign tumors) were enrolled. All panoramic radiographs were arbitrarily assigned to the training, testing 1, and testing 2 datasets of the study. The deep learning process of the training images and labels was performed for 1000 epochs using the DetectNet neural network. The testing 1 and testing 2 images were applied to the created learning model, and the detection performance was evaluated. For lesions that could be detected, the classification performance (sensitivity) for identifying radicular cysts or other lesions were examined.

RESULTS

The recall, precision, and F-1 score for detecting maxillary cysts were 74.6%/77.1%, 89.8%/90.0%, and 81.5%/83.1% for the testing 1/testing 2 datasets, respectively. The recall was higher in the anterior regions and for radicular cysts. The sensitivity was higher for identifying radicular cysts than for other lesions.

CONCLUSIONS

Using deep learning object detection technology, maxillary cyst-like lesions could be detected in approximately 75-77%.

A heuristic information cluster search approach for precise functional brain mapping

Detection of the relevant brain regions for characterizing the distinction between cognitive conditions is one of the most sought after objectives in neuroimaging research. A popular approach for achieving this goal is the multivariate pattern analysis which is currently conducted through a number of approaches such as the popular searchlight procedure. This is due to several advantages such as being automatic and flexible with regards to size of the search region. However, these approaches suffer from a number of limitations which can lead to misidentification of truly informative regions which in turn results in imprecise information maps. These limitations mainly stem from several factors such as the fact that the information value of the search spheres are assigned to the voxel at the center of them (in case of searchlight), the requirement for manual tuning of parameters such as searchlight radius and shape, and high complexity and low interpretability in commonly used machine learning-based approaches. Other drawbacks include overlooking the structure and interactions within the regions, and the disadvantages of using certain regularization techniques in analysis of datasets with characteristics of common functional magnetic resonance imaging data. In this article, we propose a fully data-driven maximum relevance minimum redundancy search algorithm for detecting precise information value of the clusters within brain regions while alleviating the above-mentioned limitations. Moreover, in order to make the proposed method faster, we propose an efficient algorithmic implementation. We evaluate and compare the proposed algorithm with the searchlight procedure as well as least absolute shrinkage and selection operator regularization-based mapping approach using both real and synthetic datasets. The analysis results of the proposed approach demonstrate higher information detection precision and map specificity compared to the benchmark approaches.

Decoding and mapping task states of the human brain via deep learning

Support vector machine (SVM)-based multivariate pattern analysis (MVPA) has delivered promising performance in decoding specific task states based on functional magnetic resonance imaging (fMRI) of the human brain. Conventionally, the SVM-MVPA requires careful feature selection/extraction according to expert knowledge. In this study, we propose a deep neural network (DNN) for directly decoding multiple brain task states from fMRI signals of the brain without any burden for feature handcrafts. We trained and tested the DNN classifier using task fMRI data from the Human Connectome Project's S1200 dataset (N = 1,034). In tests to verify its performance, the proposed classification method identified seven tasks with an average accuracy of 93.7%. We also showed the general applicability of the DNN for transfer learning to small datasets (N = 43), a situation encountered in typical neuroscience research. The proposed method achieved an average accuracy of 89.0 and 94.7% on a working memory task and a motor classification task, respectively, higher than the accuracy of 69.2 and 68.6% obtained by the SVM-MVPA. A network visualization analysis showed that the DNN automatically detected features from areas of the brain related to each task. Without incurring the burden of handcrafting the features, the proposed deep decoding method can classify brain task states highly accurately, and is a powerful tool for fMRI researchers.