Chapter 4 Local integration 2 Neural correlates of
- Slides: 25
Chapter 4: Local integration 2: Neural correlates of the BOLD signal
Overview • Introduce some of the basic principles of f. MRI • Explain how f. MRI throws up a local integration challenge • Survey some influential recent experiments on the neural correlates of the BOLD signal Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
PET • PET measures cerebral blood flow by tracking the flow of water labeled with a radioactive isotope • Basic assumption – local blood flow within the brain is related to cognitive function • Cognitive activity increased cellular activity increased blood flow • The correlation between cognitive function and blood flow has been well documented since 19 th century Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Blood flow and f. MRI • f. MRI measures levels of blood oxygenation, not blood flow • deoxygenated hemoglobin disrupts magnetic fields, while oxygenated hemoglobin does not • Levels of blood oxygenation provide an indirect measure of blood flow • oxygen consumption is not proportional to blood supply (unlike glucose) Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Blood flow and f. MRI • Cognitive activity correlated with • Increased cellular activity correlated with • Increase blood oxygen levels [because supply exceeds demand] • BOLD contrast is the contrast between oxygenated and deoxygenated blood Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Integration? • How do we move from coarse-grained correlations between blood flow and cognitive activity to an understanding of how cognitive activity takes place • We want to know not just where cognitive activity is happening, but how it is happening • Requires calibrating imaging data with data about neural activity Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Problem of levels • Neuroimaging allows us to identify which brain areas are active when subjects perform particular tasks • But there is a difference between • Localizing cognitive activity • Explaining or modeling cognitive activity Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Bridging to the neural level Brain areas • anatomically/functionally identifiable Neural networks/populations • standardly studied through computational models – behavior of populations of artificial neurons Individual neurons/small groups of neurons • can be studied through single/multi unit recordings Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Integration question • What is the neural activity that generates the BOLD contrast? • necessary first step in building neural network models • requires building bridges between different levels of organization and different technologies/tools Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Single unit recording • Using microelectrodes to investigate – how neurons respond to sensory inputs – how neurons discharge when motor acts are performed • Microelectrode recordings of interest to cognitive scientists are typically extracellular – intracellular recording very difficult in living animals Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Schematic neuron • Dendrites transmit electrostimulation from other neurons • If the combined effect of this stimulation exceeds a threshold, then the neuron generates an action potential • This action potential is transmitted via the axon Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Single unit recording • Monkey’s head held immobile • Microelectrode tip (< 10 m) inserted near neuron • can detect firing of a single neuron (action potential) • high spatial and temporal resolution Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Mirror neurons • Area F 5 of macaque monkey (premotor cortex) contains visuomotor neurons • Sensitive to different types of action (e. g. grasping vs tearing) • Some fire both when the monkey performs an action and when the monkey observes the action being performed Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
2 levels of organization Large-scale neural activity, revealed by f. MRI • ways of identifying specialization in neural areas, as a function of blood oxygen levels Fine-grained receptivity of individual neurons, as revealed in single-unit recordings The large-scale activity results from the collective activity of large numbers of individual neurons – but how? Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Neural correlate of BOLD signal Two possibilities • BOLD signal is correlated with the firing rates of populations of neurons • BOLD signal is correlated with the inputs to neurons [These are not equivalent, because neurons only fire when inputs reach a threshold] Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Rees, Friston, and Koch 2000 FMRI data on motion perception Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Calibrating with single-unit data (Rees et al. 2000) • f. MRI results show linear relationship between strength of BOLD signal in V 5 and coherence of moving stimulus • Likewise, single neurons in V 5 of macaque cortex are linearly related with motion coherence in their preferred direction • Authors propose linear relationship between strength of BOLD signal and average firing rates of neurons 9 spikes per second for each % of BOLD contrast Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Logothetis et al 2001 • Logothetis and his team measured the strength of the BOLD signal in monkey primary visual cortex at the same time as using microelectrodes to measure 2 types of neural activity • spiking activity of neurons near electrode tip • local field potentials Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Local field potential (LFP) • Electrophysiological signal representing synaptic activity at the dendrites • Corresponds to input to the neuron (and integrative processing) • Slow oscillatory wave Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Measuring LFP • LFP can be measured using the same microelectrodes as measure spiking/firing activity • Since LFP is a lower frequency signal it can be isolated through a low-pass filter • The LFP recorded at a single microelectrode represents dendritic activity in neurons within a few mm of the electrode tip Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Logothetis et al. 2000 • Anaesthetized monkey presented with rotating checkerboard pattern • Compared evolution of BOLD signal with LFP and spiking signals Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
Take home message Good news: • Logothetis experiments show to build a bridge between BOLD signal and activity of individual neurons/small populations of neurons Bad news: • The neural correlates of the BOLD signal is not the dimension of neural activity most frequently measured in single neuron studies • We don’t know much about the connection between LFP and cognition Cognitive Science José Luis Bermúdez / Cambridge University Press 2010
- Correlates of effective schools
- George‚äôs gyros
- A think local act local multicountry type of strategy
- A "think local, act local" multidomestic type of strategy
- Strategies for competing in international markets
- Forward integration and backward integration
- Xiaomi bcg matrix
- Simultaneous integration examples
- Project management chapter 4
- Chapter 7 techniques of integration
- Chapter 17 nomadic empires and eurasian integration
- Chapter 17 nomadic empires and eurasian integration
- Chapter 18 nomadic empires and eurasian integration
- Chapter 17 nomadic empires and eurasian integration
- Visualizing and understanding convolutional networks
- Vc bound
- A neural probabilistic language model
- Dan brasoveanu
- Spinal cord
- Neural theory
- Neural networks ib psychology
- Textured neural avatars
- Audio super resolution using neural networks
- Arquencéfalo
- Least mean square algorithm in neural network
- A neural image caption generator