Computational methods for analyzing wide-field calcium neuroimaging data

Zhang, Xiaohui

Permalink

https://hdl.handle.net/2142/124483 Copy

Description

Title

Computational methods for analyzing wide-field calcium neuroimaging data

Author(s)

Zhang, Xiaohui

Issue Date

2024-01-26

Director of Research (if dissertation) or Advisor (if thesis)

Anastasio, Mark A.

Doctoral Committee Chair(s)

Anastasio, Mark A.

Committee Member(s)

• Sutton, Brad P.
• Lam, Fan
• Culver, Joseph P.

Department of Study

Bioengineering

Discipline

Bioengineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Wide-field Calcium Imaging
• Functional Connectivity
• Functional Brain Networks
• Sleep State Classification
• Deep Learning
• Imaging Analysis
• Neuroimaging

Language

eng

Abstract

Wide-field calcium imaging (WFCI) with genetically encoded calcium indicators enables recordings of regional neuronal depolarization in mice across the entire cortex. Compared to traditional hemodynamics measured with functional magnetic resonance imaging (fMRI), WFCI provides a direct read-out of neural activity with higher temporal resolution and signal-to-noise ratio. Given these capabilities, WFCI has been employed to study mouse brain physiology during wakefulness, sleep, anesthesia and under disease states. While numerous efforts have been devoted to developing computational methods to analyze functional neuroimaging data such as fMRI in human in the past decades, their translation to WFCI in mice remains to be investigated. The objective of this dissertation is to develop computational methods to analyze spatiotemporal WFCI data in order to investigate two major topics of interest: identification of functional networks of the mouse brain and automated inference of sleep state. An important application in analyzing WFCI data is to understand how neuronal dynamics based on calcium signals interact with those from other brain regions by investigating the functional organization of the brain. However, traditional functional connectivity (FC) analysis generally incorporated simple bivariate Pearson correlation approach and fails to account for dependence among pixels in the rest of the brain. We first employed a multivariate functional connectivity (MFC) approach to map brain networks and impute neural activity in mice. A linear support vector regression approach was used to determine the optimal weights for integrating the signals from the remaining pixels to accurately predict neural activity in a region of interest (ROI). The resultant weight maps for each ROI were interpreted as MFC, resembled anatomical connectivity. Motivated by the recent advances of unsupervised representation learning with deep neural networks to disentangle sources from spatiotemporal neuroimaging data, we extended the use of a recurrent autoencoder to simultaneously learn spatiotemporal latent embeddings from WFCI data for identifying functional brain networks (FBNs). The resultant spatial maps of FBNs resemble those derived by traditional seed-based correlation and independent component analysis (ICA). The proposed recurrent autoencoder approach better captures subject variation and is more robust at identifying putative networks than ICA. More recently, WFCI has been employed to characterize the dynamics of neural activity during sleep. When applied to the study of sleep, WFCI data are manually scored into the different sleep states by use of adjunct electroencephalogram (EEG) and electromyograph (EMG) recordings. However, this process is time-consuming and often suffers from low inter- and intra-rater reliability and invasiveness. To better assist the sleep scoring of WFCI recording without relying on use of invasive physiology such as EEG, we developed a hybrid, two-step method by use of multiplex visibility graphs (MVGs) and two-dimensional convolutional neural network (2D CNN) for automated sleep state classification. The MVG-CNN performance is comparable with human inter-rater performance based on EEG/EMG. In order to fully explore the spatiotemporal calcium dynamics recorded by WFCI, we further extended our work to investigate an automated sleep state classifier consisting of a CNN and bidirectional long short-term memory network (BiLSTM) that jointly learn spatial and temporal information from WFCI data. The proposed approach was able to effectively distinguish among various sleep states and improve interpretability of spatial and temporal calcium dynamics exploited by the network. It is our hope that the methods proposed in this dissertation will promote broader applications of WFCI in the field of neuroscience and particularly sleep research in the future.

Graduation Semester

2024-05

Type of Resource

Text

Handle URL

https://hdl.handle.net/2142/124483

Copyright and License Information

Copyright 2024 Xiaohui Zhang

Owning Collections