FIGURE 3 - available via license: CC BY
Content may be subject to copyright.
| Confusion matrices of 1-nearest neighbor classifiers with the highest accuracy for each interactivity feature set used to identify individuals across two visits, held 1 week apart. DTW stands for dynamic time warping.
Source publication
The resting state fMRI time series appears to have cyclic patterns, which indicates presence of cyclic interactions between different brain regions. Such interactions are not easily captured by pre-established resting state functional connectivity methods including zero-lag correlation, lagged correlation, and dynamic time warping distance. These m...
Contexts in source publication
Context 1
... unlike the other feature matrices, DM demonstrated the highest stability (0.56%) when the global signal was not regressed. Figure 3 depicts the confusion matrices of 1-nearest neighbor classifiers for the conditions in which each feature matrix exhibited the highest reliability. ...
Context 2
... both stochastic and deterministic unsupervised clustering methods found a natural separation into low and high values in some two-dimensional latent space, the obtained separation was not materially related to the group labels at hand. Examples of the algorithms applied to other feature matrices can be found in Figures S3, S4. ...
Citations
... It is noninvasive, accessible, and has a high temporal resolution. It has been found 4-6 that motor movement as well as motor imagery (MI, i.e. imagination of movement without actually moving) cause modulation in SMR manifested as a decrease of power in the alpha (8)(9)(10)(11)(12)(13) Hz)/beta (13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) frequency bands, known as event-related desynchronization (ERD), followed by an increase in the beta band, also known as beta rebound or event-related synchronization (ERS), after the actual or imagined movement. Movement or MI of different body parts is associated with an SMR modulation of different regions of the sensorimotor cortex, which leads to discriminant brain signals that allows the control of MI BCI. ...
... The cyclic structure or lead-lag relationship (the temporal ordering of cyclic signals) of a multidimensional path can be recovered from the second level signature and has been applied successfully to analyse fMRI data [18][19][20] . We define the lead matrix 18 L ∈ R d×d by where S 2 (X) ∈ R d×d is the second level signature as a d × d matrix, and S 2 (X) ⊺ is its transpose. ...
Brain-computer interfaces (BCIs) allow direct communication between one’s central nervous system and a computer without any muscle movement hence by-passing the peripheral nervous system. They can restore disabled people’s ability to interact with their environment, e.g. communication and wheelchair control. However, to this day their performance is still hindered by the non-stationarity of electroencephalography (EEG) signals, as well as their susceptibility to noise from the users’ environment and from their own physiological activity. Moreover, a non-negligible amount of users struggle to use BCI systems based on motor imagery. In this paper, a new method based on the path signature is introduced to tackle this problem by using features which are different from the usual power-based ones. The path signature is a series of iterated integrals computed from a multidimensional path. It is invariant under translation and time reparametrization, which makes it a robust feature for multichannel EEG time series. The performance can be further boosted by combining the path signature with the gold standard Riemannian classifier in the BCI field exploiting the geometric structure of symmetric positive definite (SPD) matrices. The results obtained on publicly available datasets show that the signature method is more robust to inter-user variability than classical ones, especially on noisy and low-quality data. Hence, this study paves the way towards the use of mathematical tools that until now have been neglected, in order to tackle the EEG-based BCI variability issue. It also sheds light on the lead-lag relationship captured by path signature which seems relevant to assess the underlying neural mechanisms.
... The most representative and commonly used artificial technology applied to improve tinnitus therapies is machine learning [4]. For example, machine learning has been widely applied in the analysis of electroencephalogram (EEG) [5], auditory brainstem response (ABR) [6], and functional magnetic resonance imaging (fMRI) [7]. In particular, EEG can be an effective and inexpensive data source to analyze the neural feedback of tinnitus patients [8]. ...
With the development of digital technology, machine learning has paved the way for the next generation of tinnitus diagnoses. Although machine learning has been widely applied in EEG-based tinnitus analysis, most current models are dataset-specific. Each dataset may be limited to a specific range of symptoms, overall disease severity, and demographic attributes; further, dataset formats may differ, impacting model performance. This paper proposes a side-aware meta-learning for cross-dataset tinnitus diagnosis, which can effectively classify tinnitus in subjects of divergent ages and genders from different data collection processes. Owing to the superiority of meta-learning, our method does not rely on large-scale datasets like conventional deep learning models. Moreover, we design a subject-specific training process to assist the model in fitting the data pattern of different patients or healthy people. Our method achieves a high accuracy of 73.8% in the cross-dataset classification. We conduct an extensive analysis to show the effectiveness of side information of ears in enhancing model performance and side-aware meta-learning in improving the quality of the learned features.
... The most representative and commonly used artificial technology applied to improve tinnitus therapies is machine learning [3]. For example, machine learning has been widely applied in the analysis of electroencephalogram (EEG) [4], auditory brainstem response (ABR) [5], and functional magnetic resonance imaging (fMRI) [6]. In particular, EEG can be an effective and inexpensive data source to analyze the neural feedback of tinnitus patients [7]. ...
With the development of digital technology, machine learning has paved the way for the next generation of tinnitus diagnoses. Although machine learning has been widely applied in EEG-based tinnitus analysis, most current models are dataset-specific. Each dataset may be limited to a specific range of symptoms, overall disease severity, and demographic attributes; further, dataset formats may differ, impacting model performance. This paper proposes a side-aware meta-learning for cross-dataset tinnitus diagnosis, which can effectively classify tinnitus in subjects of divergent ages and genders from different data collection processes. Owing to the superiority of meta-learning, our method does not rely on large-scale datasets like conventional deep learning models. Moreover, we design a subject-specific training process to assist the model in fitting the data pattern of different patients or healthy people. Our method achieves a high accuracy of 73.8\% in the cross-dataset classification. We conduct an extensive analysis to show the effectiveness of side information of ears in enhancing model performance and side-aware meta-learning in improving the quality of the learned features.
... Global signal regression (GSR) is widely used to remove the effects of global BOLD signal variations in the analysis of fMRI studies; however, there are considerable controversies over its implementation [24][25][26], and few studies have explored the effect of GSR in ML of diagnosing neurological diseases [19,27]. Different studies have reported inconsistent results on the effect of GSR on SZ [28][29][30]. ...
Schizophrenia (SZ) is a severe psychiatric illness, and it affects around 1% of the general population; however, its reliable diagnosis is challenging. Functional MRI (fMRI) and structural MRI (sMRI) are useful techniques for investigating the functional and structural abnormalities of the human brain, and a growing number of studies have reported that multimodal brain data can improve diagnostic accuracy. Machine learning (ML) is widely used in the diagnosis of neuroscience and neuropsychiatry diseases, and it can obtain high accuracy. However, the conventional ML which concatenated the features into a longer feature vector could not be sufficiently effective to combine different features from different modalities. There are considerable controversies over the use of global signal regression (GSR), and few studies have explored the role of GSR in ML in diagnosing neurological diseases. The current study utilized fMRI and sMRI data to implement a new method named multimodal imaging and multilevel characterization with multiclassifier (M3) to classify SZs and healthy controls (HCs) and investigate the influence of GSR in SZ classification. We found that when we used Brainnetome 246 atlas and without performed GSR, our method obtained a classification accuracy of 83.49%, with a sensitivity of 68.69%, a specificity of 93.75%, and an AUC of 0.8491, respectively. We also got great classification performances with different processing methods (with/without GSR and different brain parcellation schemes). We found that the accuracy and specificity of the models without GSR were higher than that of the models with GSR. Our findings indicate that the M3 method is an effective tool to distinguish SZs from HCs, and it can identify discriminative regions to detect SZ to explore the neural mechanisms underlying SZ. The global signal may contain important neuronal information; it can improve the accuracy and specificity of SZ detection.
... In the present study, using data from the Human Connectome Project [24], we ex-panded cyclicity analysis and introduced a new technique that reveals more complex dynamics of spontaneous BOLD signals. Cyclicity analysis (CA) is a novel technique that derives pairwise temporal relations between time series using iterated path integrals (for applications in fMRI studies, see [25]). While the findings resulting from both lagged correlations and cyclicity analyses overlap and provide evidence in favor of the propagation of slow brain activity, they have different underlying mathematical apparatuses and assumptions, and levels of granularity. ...
... The lagged correlation method infers lag threads by deriving singular vectors of the time-delay matrix, whereas the CA method recovers inherent ordering among BOLD time series through eigenvectors of a lead matrix (a representation of the strength of temporal ordering between pairs of regions, see Section 2). Moreover, lagged correlation relies on interpolation and windowing to capture the dynamics of FC, which is vulnerable to time delay estimation methods, autocorrelation [25], sampling variability [26], and parameters such as window length and window shift [27]. In contrast, cyclicity analysis offers a more robust approach with a higher level of granularity to study lag structure where there is no assumption regarding stationarity, latency estimation, state duration, and state transition. ...
... Representative time course data from the HCP dataset after processing using Connectome Workbench (left) and corresponding lead matrix (right). The lead matrix is generated after an appropriate normalization of the BOLD signal (see [23,25]). ...
Fine-grained understanding of dynamics in cortical networks is crucial in unpacking brain function. Here, we introduce a novel analytical method to characterize the dynamic interaction between distant brain regions, and apply it to data from the Human Connectome Project.
Resting-state fMRI results in time series recordings of the activity of different brain regions, which are aperiodic and lacking a base frequency. Cyclicity Analysis, a novel technique robust with respect to time-reparametrizations, is effective in recovering temporal ordering of such time series along a circular trajectory without assuming any time-scale. Our analysis detected slow cortical waves of activity propagating across the brain with consistent lead-lag relationships between specific brain regions. We also observed short bursts of task-modulated strong temporal ordering that dominate overall lead-lag relationships between pairs of regions in the brain. Our results suggest the possible role played by slow waves of information transmission between brain regions that underlie emergent cognitive function.
... The fMRI data were preprocessed using Statistical Parametric Mapping software (SPM12) (http://www.fil.ion.ucl.ac.uk/spm/softw are/spm12). We employed the same preprocessing pipeline as used in our previous work (Schmidt et al., 2017;Zimmerman et al., 2018;Shahsavarani et al., 2020). To allow for magnet stabilization, the first four volumes of functional data were excluded prior to preprocessing, resulting in 300 functional volumes. ...
In the present study, we used an innovative music-rest interleaved fMRI paradigm to investigate the neural correlates of tinnitus distress. Tinnitus is a poorly-understood hearing disorder where individuals perceive sounds, in the absence of an external source. Although the great majority of individuals habituate to chronic tinnitus and report few symptoms, a minority report debilitating distress and annoyance. Prior research suggests that a diverse set of brain regions, including the attention, the salience, and the limbic networks, play key roles in mediating both the perception of tinnitus and its impact on the individual; however, evidence of the degree and extent of their involvement has been inconsistent. Here, we minimally modified the conventional resting state fMRI by interleaving it with segments of jazz music. We found that the functional connectivity between a set of brain regions–including cerebellum, precuneus, superior/middle frontal gyrus, and primary visual cortex–and seeds in the dorsal attention network, the salience network, and the amygdala, were effective in fractionating the tinnitus patients into two subgroups, characterized by the severity of tinnitus-related distress. Further, our findings revealed cross-modal modulation of the attention and salience networks by the visual modality during the music segments. On average, the more bothersome the reported tinnitus, the stronger was the exhibited inter-network functional connectivity. This study substantiates the essential role of the attention, salience, and limbic networks in tinnitus habituation, and suggests modulation of the attention and salience networks across the auditory and visual modalities as a possible compensatory mechanism for bothersome tinnitus.
Background: Resting state functional Magnetic Resonance Imaging (rsfMRI) differences have been reported in individuals with anorexia nervosa (AN). However, methodological issues limit inferences, the dynamic cyclic nature of the resting state signal has not been explored, neither has the association with known elevated autistic characteristics.
Methods: 92 participants, 65 Individuals with AN and 27 controls underwent rsfMRI. Cyclic analysis was conducted to obtain pairwise relationships between regions. Classification of group and regression to predict autistic characteristics within the AN group was conducted, with model weights being explored to ascertain the most predictive pairwise relationships.
Results: Pairwise relationships in the temporal, dorsal and ventral attention networks were most predictive of group. The anterior intraparietal sulcus, salience, dorsal attention, auditory dorsal posterior cingulate cortex, and ventral attention networks were most predictive of autistic characteristics.
Discussion: Several distinct pairwise relationships predicting group and autistic characteristics were found, however, the global disruption of the temporal ordering of the cyclic wave, and variation in temporal ordering across resting state scan are also a neurophenotype in individuals with AN and the relationship to autistic characteristics. Characteristics associated with AN and autism are also predicted by distinct neural regions.
Resting state functional connectivity (RS-FC) studies of tinnitus over the years have produced inconsistent results. While findings can be organized into broad categories, such as increased correlations between auditory and limbic areas in tinnitus patients and a disrupted default mode network, there has been little one-to-one correspondence of results across RS-FC studies of tinnitus. While some of this variation can be explained by the heterogeneity of the tinnitus population, including tinnitus severity, the sources of variability in RS-FC of tinnitus patients are unclear. To directly assess the reliability of RS-FC measures in tinnitus, both tinnitus and control participants from two different sites (University of Illinois at Urbana-Champaign, or UIUC, and the Wilford Hall Ambulatory Surgical Center, or WHASC, at the Lackland Airforce Base in San Antonio, Texas) participated in two resting state MRI scans separated by exactly one week. Seed-to-seed analysis assessing correlations between the fMRI activity of 27 regions in the default mode, dorsal attention, auditory, visual, salience, and emotional processing networks were examined in control and tinnitus participants separately for each site. Additionally, heart rate and respiration measures were collected at UIUC, and the effect of extra physiological corrections using these measures on reliability was examined within the UIUC participants. Intra-class correlation coefficients (ICCs) were used as the measure of reliability. Overall, RS-FC in a seed-to-seed analysis was as reliable in tinnitus participants as it was in control participants in the seed regions examined. As previously shown in studies of participants with normal hearing sensitivity, intra-network reliability was higher than inter-network reliability. Related to this, stronger correlations between two seed regions were predictive of stronger reliability of the connectivity between those regions. These effects were seen in both control and tinnitus populations. Additional physiological corrections did not have a significant impact on the ICC values. The current study demonstrates that, on a whole-brain level, RS-FC assessed via seed-to-seed analysis is reliable in tinnitus participants. We therefore must look to other sources as potential causes of discrepancies across studies, such as variability within analysis techniques or within the behavioral characteristics of tinnitus participants.
Time-varying phenomena are ubiquitous across pure and applied mathematics, from path spaces and stochastic differential equations to multivariate time series and dynamic point clouds. The path signature provides a powerful characterization of such sequential data in terms of power series of tensors, weaving together these diverse concepts. Originally defined as part of Chen's iterated integral cochain algebra, the path signature has since been used as the foundation for the theory of rough paths in stochastic analysis. More recently, it has been shown to be a universal and characteristic feature map for multivariate time series, providing theoretical guarantees for its application to time series analysis in the context of kernel methods in machine learning. This thesis extends the scope of the path signature to more complex parametrized data in two directions. First, we consider generalizations of the codomain of a path. We lift the theory of signatures to the setting of Lie group valued time series, adapting these tools for time series with underlying geometric constraints. Furthermore, we build a signature framework to study paths of persistence diagrams, objects which capture the evolving topological structure of dynamic data sets. Second, we consider maps parametrized by higher dimensional cubes by developing notions of the mapping space signature. Our approach returns to the topological origins of the signature as the 0-cochains of the Chen construction. We formulate a cubical variant of the mapping space construction, and use the resulting 0-cochains to define the mapping space signature and establish its basic properties.
