Wavelet correlation between subjects: A time-scale data driven analysis for brain mapping using fMRI

A CNN Deep Local and Global ASD Classification Approach with Continuous Wavelet Transform Using Task-Based FMRI

Autism spectrum disorder (ASD) is a neurodegenerative disorder characterized by lingual and social disabilities. The autism diagnostic observation schedule is the current gold standard for ASD diagnosis. Developing objective computer aided technologies for ASD diagnosis with the utilization of brain imaging modalities and machine learning is one of main tracks in current studies to understand autism. Task-based fMRI demonstrates the functional activation in the brain by measuring blood oxygen level-dependent (BOLD) variations in response to certain tasks. It is believed to hold discriminant features for autism. A novel computer aided diagnosis (CAD) framework is proposed to classify 50 ASD and 50 typically developed toddlers with the adoption of CNN deep networks. The CAD system includes both local and global diagnosis in a response to speech task. Spatial dimensionality reduction with region of interest selection and clustering has been utilized. In addition, the proposed framework performs discriminant feature extraction with continuous wavelet transform. Local diagnosis on cingulate gyri, superior temporal gyrus, primary auditory cortex and angular gyrus achieves accuracies ranging between 71% and 80% with a four-fold cross validation technique. The fused global diagnosis achieves an accuracy of 86% with 82% sensitivity, 92% specificity. A brain map indicating ASD severity level for each brain area is created, which contributes to personalized diagnosis and treatment plans.

A robust DWT-CNN-based CAD system for early diagnosis of autism using task-based fMRI

PURPOSE

Task-based fMRI (TfMRI) is a diagnostic imaging modality for observing the effects of a disease or other condition on the functional activity of the brain. Autism spectrum disorder (ASD) is a pervasive developmental disorder associated with impairments in social and linguistic abilities. Machine learning algorithms have been widely utilized for brain imaging aiming for objective ASD diagnostics. Recently, deep learning methods have been gaining more attention for fMRI classification. The goal of this paper is to develop a convolutional neural network (CNN)-based framework to help in global diagnosis of ASD using TfMRI data that are collected from a response to speech experiment.

METHODS

To achieve this goal, the proposed framework adopts a novel imaging marker integrating both spatial and temporal information that are related to the functional activity of the brain. The developed pipeline consists of three main components. In the first step, the collected TfMRI data are preprocessed and parcellated using the Harvard-Oxford probabilistic atlas included with the fMRIB Software Library (FSL). Second, a group analysis using FSL is performed between ASD and typically developing (TD) children to identify significantly activated brain areas in response to the speech task. In order to reduce brain spatial dimensionality, a K-means clustering technique is performed on such significant brain areas. Informative blood oxygen level-dependent (BOLD) signals are extracted from each cluster. A compression step for each extracted BOLD signal using discrete wavelet transform (DWT) has been proposed. The adopted wavelets are similar to the expected hemodynamic response which enables DWT to compress the BOLD signal while highlighting its activation information. Finally, a deep learning 2D CNN network is used to classify the patients as ASD or TD based on extracted features from the previous step.

RESULTS

Preliminary results on 100 TfMRI dataset (50 ASD, 50 TD) obtain 80% correct global classification using tenfold cross validation (with sensitivity = 84%, specificity = 76%).

CONCLUSION

The experimental results show the high accuracy of the proposed framework and hold promise for the presented framework as a helpful adjunct to currently used ASD diagnostic tools.

Combining the intersubject correlation analysis and the multivariate distance matrix regression to evaluate associations between fNIRS signals and behavioral data from ecological experiments

The development of methods to analyze data acquired using functional near-infrared spectroscopy (fNIRS) in experiments similar to real-life situations is of great value in modern applied neuroscience. One of the most used methods to analyze fNIRS signals consists of the application of the general linear model on the observed hemodynamic signals. However, it implies limitations on the experimental design that must be constrained by triggers related to the stimuli protocols (such as block design or event related). In this work, a novel methodology is proposed to overcome such restrictions and allow more flexible protocols. The method combines the intersubject correlation analysis and the multivariate distance matrix regression to evaluate the brain-behavior relationship of subjects submitted to experiments with no trigger-based protocols. Its applicability is demonstrated throughout a naturalistic experiment about emotions conveyed by music. Thirty-two participants freely listened to instrumental excerpts from the operatic repertoire and reported the valences of the emotions conveyed by the musical segments. The method was able to find a statistically significant correlation between the subjects' fNIRS signals and valences of their emotional responses, for the excerpt that evoked the most negative valence. This result illustrates the potential of this approach as an alternative method to analyze fNIRS signals from experiments in which block design or task-related paradigms might not be suitable.

Cortical Response Similarities Predict which Audiovisual Clips Individuals Viewed, but Are Unrelated to Clip Preference

Cortical responses to complex natural stimuli can be isolated by examining the relationship between neural measures obtained while multiple individuals view the same stimuli. These inter-subject correlation’s (ISC’s) emerge from similarities in individual’s cortical response to the shared audiovisual inputs, which may be related to their emergent cognitive and perceptual experience. Within the present study, our goal is to examine the utility of using ISC’s for predicting which audiovisual clips individuals viewed, and to examine the relationship between neural responses to natural stimuli and subjective reports. The ability to predict which clips individuals viewed depends on the relationship of the EEG response across subjects and the nature in which this information is aggregated. We conceived of three approaches for aggregating responses, i.e. three assignment algorithms, which we evaluated in Experiment 1A. The aggregate correlations algorithm generated the highest assignment accuracy (70.83% chance = 33.33%) and was selected as the assignment algorithm for the larger sample of individuals and clips within Experiment 1B. The overall assignment accuracy was 33.46% within Experiment 1B (chance = 06.25%), with accuracies ranging from 52.9% (Silver Linings Playbook) to 11.75% (Seinfeld) within individual clips. ISC’s were significantly greater than zero for 15 out of 16 clips, and fluctuations within the delta frequency band (i.e. 0-4 Hz) primarily contributed to response similarities across subjects. Interestingly, there was insufficient evidence to indicate that individuals with greater similarities in clip preference demonstrate greater similarities in cortical responses, suggesting a lack of association between ISC and clip preference. Overall these results demonstrate the utility of using ISC’s for prediction, and further characterize the relationship between ISC magnitudes and subjective reports.

Uni- and multisensory brain areas are synchronised across spectators when watching unedited dance recordings

The superior temporal sulcus (STS) and gyrus (STG) are commonly identified to be functionally relevant for multisensory integration of audiovisual (AV) stimuli. However, most neuroimaging studies on AV integration used stimuli of short duration in explicit evaluative tasks. Importantly though, many of our AV experiences are of a long duration and ambiguous. It is unclear if the enhanced activity in audio, visual, and AV brain areas would also be synchronised over time across subjects when they are exposed to such multisensory stimuli. We used intersubject correlation to investigate which brain areas are synchronised across novices for uni- and multisensory versions of a 6-min 26-s recording of an unfamiliar, unedited Indian dance recording (Bharatanatyam). In Bharatanatyam, music and dance are choreographed together in a highly intermodal-dependent manner. Activity in the middle and posterior STG was significantly correlated between subjects and showed also significant enhancement for AV integration when the functional magnetic resonance signals were contrasted against each other using a general linear model conjunction analysis. These results extend previous studies by showing an intermediate step of synchronisation for novices: while there was a consensus across subjects' brain activity in areas relevant for unisensory processing and AV integration of related audio and visual stimuli, we found no evidence for synchronisation of higher level cognitive processes, suggesting these were idiosyncratic.

Sensory processing during viewing of cinematographic material: computational modeling and functional neuroimaging

The investigation of brain activity using naturalistic, ecologically-valid stimuli is becoming an important challenge for neuroscience research. Several approaches have been proposed, primarily relying on data-driven methods (e.g. independent component analysis, ICA). However, data-driven methods often require some post-hoc interpretation of the imaging results to draw inferences about the underlying sensory, motor or cognitive functions. Here, we propose using a biologically-plausible computational model to extract (multi-)sensory stimulus statistics that can be used for standard hypothesis-driven analyses (general linear model, GLM). We ran two separate fMRI experiments, which both involved subjects watching an episode of a TV-series. In Exp 1, we manipulated the presentation by switching on-and-off color, motion and/or sound at variable intervals, whereas in Exp 2, the video was played in the original version, with all the consequent continuous changes of the different sensory features intact. Both for vision and audition, we extracted stimulus statistics corresponding to spatial and temporal discontinuities of low-level features, as well as a combined measure related to the overall stimulus saliency. Results showed that activity in occipital visual cortex and the superior temporal auditory cortex co-varied with changes of low-level features. Visual saliency was found to further boost activity in extra-striate visual cortex plus posterior parietal cortex, while auditory saliency was found to enhance activity in the superior temporal cortex. Data-driven ICA analyses of the same datasets also identified "sensory" networks comprising visual and auditory areas, but without providing specific information about the possible underlying processes, e.g., these processes could relate to modality, stimulus features and/or saliency. We conclude that the combination of computational modeling and GLM enables the tracking of the impact of bottom-up signals on brain activity during viewing of complex and dynamic multisensory stimuli, beyond the capability of purely data-driven approaches.