310 likes | 430 Vues
This work explores the evolution of neurophysiological signals, focusing on data characteristics such as small signal-to-noise ratios and nonstationary time series. It examines evolving replicates across experiments, particularly in a learning association framework. Models such as weakly stationary and locally stationary time series are analyzed. The study emphasizes the importance of understanding frequency power differences in familiar versus novel trials, leveraging the Log Periodogram methods. It highlights the significance of advanced statistical techniques for neuroimaging data to address complex scientific questions.
E N D
Modeling the Evolution of Neurophysiological Signals Mark Fiecas Hernando Ombao
Data Characteristics Small signal-to-noise ratios
Data Characteristics Nonstationarytime series data
Data Characteristics Evolving over time within a replicate Nonidentical replicates across the experiment
The Time Series Models Weakly stationary time series (Brillinger, 1981):
The Time Series Models Locally stationary time series (Dahlhaus, 2000):
The Time Series Models Locally stationary time series with slowly evolving replicates:
The Time Series Models • Replicates are uncorrelated. For each replicate, use existing methods to address nonstationarity over time. • Smooth the estimates over time and replicate-time.
A Relevant Scientific Question Is the power in a frequency band of interest the same between “familiar” and “novel” trials?
Log Periodogram Models Weakly stationary data (Krafty et al, 2011):
Log Periodogram Models Weakly stationary data (Krafty et al, 2011):where
The Log Periodogram Models Locally stationary data (Krafty, 2007; Qin and Guo, 2009):
The Log Periodogram Models Locally stationary data (Krafty et al, 2007):where
Calling All Statisticians “Understanding how the brain works is arguably one of the greatest scientific challenges of our time.” - Alivisatos et al, 2013
Calling All Statisticians • The BRAIN Initiative (USA) • The Human Brain Project (European Union) • 86 Institutions in Europe involved • €1 billion in funding / year
Calling All Statisticians Very rich data sets • High temporal resolution (EEG, MEG, LFP) • High spatial resolution (PET, fMRI) • 300k spatial locations in fMRI • Imaging genetics Many open problems
Calling All Statisticians Handbook of Modern Statistical Methods: Neuroimaging Data Analysis (eds: H. Ombao, M. Lindquist, W. Thompson, and J. Aston)
Acknowledgments • Shaun Patel, Boston University • EmadEskandar, MGH