Pinzuti, Edoardo (2023)
Measuring spectrally resolved information processing in neural data.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00024780
Dissertation, Erstveröffentlichung, Verlagsversion
Kurzbeschreibung (Abstract)
Background:
The human brain, an incredibly complex biological system comprising billions of neurons and trillions of synapses, possesses remarkable capabilities for information processing and distributed computations. Neurons, the fundamental building blocks, perform elementary operations on their inputs and collaborate extensively to execute intricate computations, giving rise to cognitive functions and behavior. Notably, distributed information processing in the brain heavily relies on rhythmic neural activity characterized by synchronized oscillations at specific frequencies. These oscillations play a crucial role in coordinating brain activity and facilitating communication between different neural circuits [1], effectively acting as temporal windows that enable efficient information exchange within specific frequency ranges. To understand distributed information processing in neural systems, breaking down its components, i.e., —information transfer, storage, and modification can be helpful, but requires precise mathematical definitions for each respective component. Thankfully, these definitions have recently become available [2]. Information theory is a natural choice for measuring information processing, as it offers a mathematically complete description of the concept of information and communication. The fundamental information-processing operations, are considered essential prerequisites for achieving universal information processing in any system [3]. By quantifying and analyzing these operations, we gain valuable insights into the brain’s complex computation and cognitive abilities. As information processing in the brain is intricately tied to rhythmic behavior, there is a need to establish a connection between information theoretic measures and frequency components. Previous attempts to achieve frequency-resolved information theoretic measures have mostly relied on narrowband filtering [4], which comes with several known issues of phase shifting and high false positive rate results [5], or simplifying the computation to few variables [6], that might result in missing important information in the analysed brain signals. Therefore, the current work aims to establish a frequency-resolved measure of two crucial components of information processing: information transfer and information storage. By proposing methodological advancements, this research seeks to shed light on the role of neural oscillations in information processing within the brain. Furthermore, a more comprehensive investigation was carried out on the communication between two critical brain regions responsible for motor inhibition in the frontal cortex (right Inferior Frontal gyrus (rIFG) and pre-Supplementary motor cortex (pre-SMA)). Here, neural oscillations in the beta band (12 − 30 Hz) have been proposed to have a pivotal role in response inhibition. A long-standing question in the field was to disentangle which of these two brain areas first signals the stopping process and drives the other [7]. Furthermore, it was hypothesized that beta oscillations carry the information transfer between these regions. The present work addresses the methodological problems and investigates spectral information processing in neural data, in three studies. Study 1 focuses on the critical role of information transfer, measured by transfer entropy, in distributed computation. Understanding the patterns of information transfer is essential for unraveling the computational algorithms in complex systems, such as the brain. As many natural systems rely on rhythmic processes for distributed computations, a frequency-resolved measure of information transfer becomes highly valuable. To address this, a novel algorithm is presented, efficiently identifying frequencies responsible for sending and receiving information in a network. The approach utilizes the invertible maximum overlap discrete wavelet transform (MODWT) to create surrogate data for computing transfer entropy, eliminating issues associated with phase shifts and filtering. However, measuring frequency-resolved information transfer poses a Partial information decomposition problem [8] that is yet to be fully resolved. The algorithm’s performance is validated using simulated data and applied to human magnetoencephalography (MEG) and ferret local field potential recordings (LFP). In human MEG, the study unveils a complex spectral configuration of cortical information transmission, showing top-down information flow from very high frequencies (above 100Hz) to both similarly high frequencies and frequencies around 20Hz in the temporal cortex. Contrary to the current assumption, the findings suggest that low frequencies do not solely send information to high frequencies. In the ferret LFP, the prefrontal cortex demonstrates the transmission of information at low frequencies, specifically within the range of 4-8 Hz. On the receiving end, V1 exhibits a preference for operating at very high frequency > 125 Hz. The spectrally resolved transfer entropy promises to deepen our understanding of rhythmic information exchange in natural systems, shedding light on the computational properties of oscillations on cognitive functions. In study 2, the primary focus lay on the second fundamental aspect of information processing: the active information storage (AIS). The AIS estimates how much information in the next measurements of the process can be predicted by examining its paste state. In processes that either produce little information (low entropy) or that are highly unpredictable, the AIS is low, whereas processes that are predictable but visit many different states with equal probabilities, exhibit high AIS [9]. Within this context, we introduced a novel spectrally-resolved AIS. Utilizing intracortical recordings of neural activity in anesthetized ferrets before and after loss of consciousness (LOC), the study reveals that the modulation of AIS by anesthesia is highly specific to different frequency bands, cortical layers, and brain regions. The findings reveal that the effects of anesthesia on AIS are prominent in the supragranular layers for the high/low gamma band, while the alpha/beta band exhibits the strongest decrease in AIS at infragranular layers, in accordance with the predictive coding theory. Additionally, the isoflurane impacts local information processing in a frequency-specific manner. For instance, increases in isoflurane concentration lead to a decrease in AIS in the alpha frequency but to an increase in AIS in the delta frequency range (<2Hz). In sum, analyzing spectrally-resolved AIS provides valuable insights into changes in cortical information processing under anesthesia. With rhythmic neural activity playing a significant role in biological neural systems, the introduction of frequency-specific components in active information storage allows a deeper understanding of local information processing in different brain areas and under various conditions. In study 3, to further verify the pivotal role of neural oscillations in information processing, we investigated the neural network mechanisms underlying response inhibition. A long-standing debate has centered around identifying the cortical initiator of response inhibition in the beta band, with two main regions proposed: the right rIFG and the pre-SMA. This third study aimed to determine which of these regions is activated first and exerts a potential information exchange on the other. Using high temporal resolution magnetoencephalography (MEG) and a relatively large cohort of subjects. A significant breakthrough is achieved by demonstrating that the rIFG is activated significantly earlier than the pre-SMA. The onset of beta band activity in the rIFG occurred at around 140 ms after the STOP signal. Further analyses showed that the beta-band activity in the rIFG was crucial for successful stopping, as evidenced by its predictive value for stopping performance. Connectivity analysis revealed that the rIFG sends information in the beta band to the pre-SMA but not vice versa, emphasizing the rIFG’s dominance in the response inhibition process. The results provide strong support for the hypothesis that the rIFG initiates stopping and utilizes beta-band oscillations for this purpose. These findings have significant implications, suggesting the possibility of spatially localized oscillation based interventions for response inhibition.
Conclusion:
In conclusion, the present work proposes a novel algorithm for uncovering the frequencies at which information is transferred between sources and targets in the brain, providing valuable insights into the computational dynamics of neural processes. The spectrally resolved transfer entropy was successfully applied to experimental neural data of intracranial recordings in ferrets and MEG recordings of humans. Furthermore, the study on active information storage (AIS) analysis under anesthesia revealed that the spectrally resolved AIS offers unique additional insights beyond traditional spectral power analysis. By examining changes in neural information processing, the study demonstrates how AIS analysis can deepen the understanding of anesthesia’s effects on cortical information processing. Moreover, the third study’s findings provide strong evidence supporting the critical role of beta oscillations in information processing, particularly in response inhibition. The research successfully demonstrates that beta oscillations in the rIFG functions as the key initiator of the response inhibition process, acting as a top-down control mechanism. The identification of beta oscillations as a crucial factor in information processing opens possibilities for further research and targeted interventions in neurological disorders. Taken together, the current work highlights the role of spectrally-resolved information processing in neural systems by not only introducing novel algorithms, but also successfully applying them to experimental oscillatory neural activity in relation to low-level cortical information processing (anesthesia) as well as high-level processes (cognitive response inhibition).
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2023 | ||||
Autor(en): | Pinzuti, Edoardo | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Measuring spectrally resolved information processing in neural data | ||||
Sprache: | Englisch | ||||
Referenten: | Galuske, Prof. Dr. Ralf ; Wibral, Prof. Dr. Michael | ||||
Publikationsjahr: | 15 November 2023 | ||||
Ort: | Darmstadt | ||||
Kollation: | xviii, 216 Seiten | ||||
Datum der mündlichen Prüfung: | 6 November 2023 | ||||
DOI: | 10.26083/tuprints-00024780 | ||||
URL / URN: | https://tuprints.ulb.tu-darmstadt.de/24780 | ||||
Kurzbeschreibung (Abstract): | Background: The human brain, an incredibly complex biological system comprising billions of neurons and trillions of synapses, possesses remarkable capabilities for information processing and distributed computations. Neurons, the fundamental building blocks, perform elementary operations on their inputs and collaborate extensively to execute intricate computations, giving rise to cognitive functions and behavior. Notably, distributed information processing in the brain heavily relies on rhythmic neural activity characterized by synchronized oscillations at specific frequencies. These oscillations play a crucial role in coordinating brain activity and facilitating communication between different neural circuits [1], effectively acting as temporal windows that enable efficient information exchange within specific frequency ranges. To understand distributed information processing in neural systems, breaking down its components, i.e., —information transfer, storage, and modification can be helpful, but requires precise mathematical definitions for each respective component. Thankfully, these definitions have recently become available [2]. Information theory is a natural choice for measuring information processing, as it offers a mathematically complete description of the concept of information and communication. The fundamental information-processing operations, are considered essential prerequisites for achieving universal information processing in any system [3]. By quantifying and analyzing these operations, we gain valuable insights into the brain’s complex computation and cognitive abilities. As information processing in the brain is intricately tied to rhythmic behavior, there is a need to establish a connection between information theoretic measures and frequency components. Previous attempts to achieve frequency-resolved information theoretic measures have mostly relied on narrowband filtering [4], which comes with several known issues of phase shifting and high false positive rate results [5], or simplifying the computation to few variables [6], that might result in missing important information in the analysed brain signals. Therefore, the current work aims to establish a frequency-resolved measure of two crucial components of information processing: information transfer and information storage. By proposing methodological advancements, this research seeks to shed light on the role of neural oscillations in information processing within the brain. Furthermore, a more comprehensive investigation was carried out on the communication between two critical brain regions responsible for motor inhibition in the frontal cortex (right Inferior Frontal gyrus (rIFG) and pre-Supplementary motor cortex (pre-SMA)). Here, neural oscillations in the beta band (12 − 30 Hz) have been proposed to have a pivotal role in response inhibition. A long-standing question in the field was to disentangle which of these two brain areas first signals the stopping process and drives the other [7]. Furthermore, it was hypothesized that beta oscillations carry the information transfer between these regions. The present work addresses the methodological problems and investigates spectral information processing in neural data, in three studies. Study 1 focuses on the critical role of information transfer, measured by transfer entropy, in distributed computation. Understanding the patterns of information transfer is essential for unraveling the computational algorithms in complex systems, such as the brain. As many natural systems rely on rhythmic processes for distributed computations, a frequency-resolved measure of information transfer becomes highly valuable. To address this, a novel algorithm is presented, efficiently identifying frequencies responsible for sending and receiving information in a network. The approach utilizes the invertible maximum overlap discrete wavelet transform (MODWT) to create surrogate data for computing transfer entropy, eliminating issues associated with phase shifts and filtering. However, measuring frequency-resolved information transfer poses a Partial information decomposition problem [8] that is yet to be fully resolved. The algorithm’s performance is validated using simulated data and applied to human magnetoencephalography (MEG) and ferret local field potential recordings (LFP). In human MEG, the study unveils a complex spectral configuration of cortical information transmission, showing top-down information flow from very high frequencies (above 100Hz) to both similarly high frequencies and frequencies around 20Hz in the temporal cortex. Contrary to the current assumption, the findings suggest that low frequencies do not solely send information to high frequencies. In the ferret LFP, the prefrontal cortex demonstrates the transmission of information at low frequencies, specifically within the range of 4-8 Hz. On the receiving end, V1 exhibits a preference for operating at very high frequency > 125 Hz. The spectrally resolved transfer entropy promises to deepen our understanding of rhythmic information exchange in natural systems, shedding light on the computational properties of oscillations on cognitive functions. In study 2, the primary focus lay on the second fundamental aspect of information processing: the active information storage (AIS). The AIS estimates how much information in the next measurements of the process can be predicted by examining its paste state. In processes that either produce little information (low entropy) or that are highly unpredictable, the AIS is low, whereas processes that are predictable but visit many different states with equal probabilities, exhibit high AIS [9]. Within this context, we introduced a novel spectrally-resolved AIS. Utilizing intracortical recordings of neural activity in anesthetized ferrets before and after loss of consciousness (LOC), the study reveals that the modulation of AIS by anesthesia is highly specific to different frequency bands, cortical layers, and brain regions. The findings reveal that the effects of anesthesia on AIS are prominent in the supragranular layers for the high/low gamma band, while the alpha/beta band exhibits the strongest decrease in AIS at infragranular layers, in accordance with the predictive coding theory. Additionally, the isoflurane impacts local information processing in a frequency-specific manner. For instance, increases in isoflurane concentration lead to a decrease in AIS in the alpha frequency but to an increase in AIS in the delta frequency range (<2Hz). In sum, analyzing spectrally-resolved AIS provides valuable insights into changes in cortical information processing under anesthesia. With rhythmic neural activity playing a significant role in biological neural systems, the introduction of frequency-specific components in active information storage allows a deeper understanding of local information processing in different brain areas and under various conditions. In study 3, to further verify the pivotal role of neural oscillations in information processing, we investigated the neural network mechanisms underlying response inhibition. A long-standing debate has centered around identifying the cortical initiator of response inhibition in the beta band, with two main regions proposed: the right rIFG and the pre-SMA. This third study aimed to determine which of these regions is activated first and exerts a potential information exchange on the other. Using high temporal resolution magnetoencephalography (MEG) and a relatively large cohort of subjects. A significant breakthrough is achieved by demonstrating that the rIFG is activated significantly earlier than the pre-SMA. The onset of beta band activity in the rIFG occurred at around 140 ms after the STOP signal. Further analyses showed that the beta-band activity in the rIFG was crucial for successful stopping, as evidenced by its predictive value for stopping performance. Connectivity analysis revealed that the rIFG sends information in the beta band to the pre-SMA but not vice versa, emphasizing the rIFG’s dominance in the response inhibition process. The results provide strong support for the hypothesis that the rIFG initiates stopping and utilizes beta-band oscillations for this purpose. These findings have significant implications, suggesting the possibility of spatially localized oscillation based interventions for response inhibition. Conclusion: In conclusion, the present work proposes a novel algorithm for uncovering the frequencies at which information is transferred between sources and targets in the brain, providing valuable insights into the computational dynamics of neural processes. The spectrally resolved transfer entropy was successfully applied to experimental neural data of intracranial recordings in ferrets and MEG recordings of humans. Furthermore, the study on active information storage (AIS) analysis under anesthesia revealed that the spectrally resolved AIS offers unique additional insights beyond traditional spectral power analysis. By examining changes in neural information processing, the study demonstrates how AIS analysis can deepen the understanding of anesthesia’s effects on cortical information processing. Moreover, the third study’s findings provide strong evidence supporting the critical role of beta oscillations in information processing, particularly in response inhibition. The research successfully demonstrates that beta oscillations in the rIFG functions as the key initiator of the response inhibition process, acting as a top-down control mechanism. The identification of beta oscillations as a crucial factor in information processing opens possibilities for further research and targeted interventions in neurological disorders. Taken together, the current work highlights the role of spectrally-resolved information processing in neural systems by not only introducing novel algorithms, but also successfully applying them to experimental oscillatory neural activity in relation to low-level cortical information processing (anesthesia) as well as high-level processes (cognitive response inhibition). |
||||
Alternatives oder übersetztes Abstract: |
|
||||
Status: | Verlagsversion | ||||
URN: | urn:nbn:de:tuda-tuprints-247807 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 000 Allgemeines, Informatik, Informationswissenschaft > 004 Informatik 500 Naturwissenschaften und Mathematik > 500 Naturwissenschaften 500 Naturwissenschaften und Mathematik > 570 Biowissenschaften, Biologie |
||||
Fachbereich(e)/-gebiet(e): | 10 Fachbereich Biologie 10 Fachbereich Biologie > Neurophysiologie und neurosensorische Systeme |
||||
Hinterlegungsdatum: | 15 Nov 2023 10:44 | ||||
Letzte Änderung: | 16 Nov 2023 06:06 | ||||
PPN: | |||||
Referenten: | Galuske, Prof. Dr. Ralf ; Wibral, Prof. Dr. Michael | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 6 November 2023 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |