HHB2003 - Neuro Methods 2 2024 PDF

HHB2003 - Neuroscience of cognition: Neuroscience Methods & Issues -2 Dr Chris Retzler ([email protected]) Focus on following areas and what each can tell us about cognitive function: 1. Neural function (EEG, MEG) 2. Combined measures of function and structure (fMRI, PET) 3. General issue...

HHB2003 - Neuroscience of cognition: Neuroscience Methods & Issues -2 Dr Chris Retzler ([email protected]) Focus on following areas and what each can tell us about cognitive function: 1. Neural function (EEG, MEG) 2. Combined measures of function and structure (fMRI, PET) 3. General issues with neuro methods 4. Are combined methods the future? Functional neuroimaging methods: temporal and spatial resolution Meyer-Lindenberg. 2010. Nature 1. Neural function (EEG, MEG) Electroencephalography (EEG) Measures large populations of active neurons (summation) Continuous measure EEG varies according to behavioural state Can be used to assess abnormalities, e.g. epilepsy EEG during epilepsy Fig. 1. An example of the EEG recording of epileptic seizure. Paroxysmal 3 Hz spike and wave pattern emerges abruptly out of normal background and suddenly ceases after few seconds. Lopes da Silva et al. (2003) EEG pre- processing: ARTIFACT REJECTION Blinks, movements etc. FILTER EPOCH Reject bad channels & interpolate Reject epochs BASELINE CORRECT AVERAGE ALTERNATIVE (FOR ERPs) ANALYSIS Noisy data! Eye flutter Muscle artifacts Movements ??? EEG artifact Raw EEG before blink rejection correction: Visual inspection Thresholding (extreme values) Raw EEG after blink rejection PCA/ICA Event-Related Potential (ERP) technique: How is brain activity modulated by a cognitive task? Average across trials of a particular type (time locked) This removes variations in the EEG which are not related to the event of interest Watch out for ERPs plotted negative up P/N Figure 1. Schematic overview of the oddball paradigm and an example of an ERP. Why might this signal be important? van Dinteren et al. (2014) Frequency analysis of EEG data Synchronized firing of neurons produces oscillatory activity which is defined by the frequency of this activity (e.g. 10 Hz = 10 times per second) Time-frequency analysis refers to the changing amplitude (power) of a wave across time. Event-related Delta: 0.1 or resting data – 3.5 Hz Theta: 4.0 – 7.5 Hz Alpha: 8.0 – 13 Hz Beta: 14.0 – 30.0 Hz Gamma: 30.0 – 100.0 Hz Example: the N400 ERP However interpretation can be more difficult… When the N400 was first discovered, many linguists were excited, because it seemed that this could be the brain activity pattern that reveals how the brain understands sentences However the N400 may not be related to sentence understanding… Instead it may be due to priming: cream…sugar vs. cream…dog This would simply reflect how related a word is to other words EEG issues and considerations  Excellent temporal (when) resolution (ms)  Non-invasive  Relatively low cost  Can be more sensitive than behavioural paradigms X Poor spatial (where) resolution X Long set-up time X Low SNR Magnetoencephalography (MEG): Records magnetic fields generated by neuronal sources of the brain These fields are detected by super- conducting detectors and amplifiers known as TheSQUIDS fields are analysed to identify the neuronal sources within the brain Overlay images on MRI Barnes et al. (2010), Scholarpedia  MEG analysis: Auditory N1m (A) (B) (C) (D) (A)SQUIDS positioned around head (B)Subject hears a tone which elicits an auditory N1m (C)Contour maps at 100 ms (D)Dipole localized in the auditory area of temporal cortex Aims: To maximise on the spatial and temporal resolution of MEG (based on previous EEG work) Looking at recognition memory Method: Word recognition task – Train: Incorporate words into a sentence – Test: “Have you seen this word?” MEG analysis looks at magnetic evoked fields (MEFs) and dipole fitting (source location) Tendolkar et al. (2000). Neuroscience Results: Compare recognised words and new words MEFs to correctly recognised words were different between 400- 1000ms This magnetic old/new effect was more widespread over the left than the right hemisphere Tendolkar et al. (2000). Neuroscience Summary: MEFs revealed a difference between studied and unstudied stimuli. MEG yielded reliable source information revealing the activity of three independent dipoles, located in the right medial temporal lobe (MTL), the right inferior frontal and the left inferior parietal cortices. Building on EEG results, MEG identified latency of differences and location So MEG may have an important role in finding the neural correlates of fast-acting cognitive processes. Tendolkar et al. (2000). Neuroscience Task: Write down one advantage and one disadvantage of MEG as a neuroscience technique MEG issues and considerations: X Signals of interest are very small X Inverse problem (determine the location of electric activity within the brain from the magnetic fields outside the head)  Very good temporal resolution  Direct measure of brain function (unlike fMRI and PET)  Non-invasive  Good spatial resolution (better than EEG at least) 2. Structural analysis – Magnetic Resonance Imaging (MRI) fMRI intro video here  Magnetic Resonance Imaging (MRI): (A) (B) (C) (D) MRI exploits the magnetic properties of organic tissue (e.g. hydrogen atoms) in the brain. (A)In normal state, orientation of these hydrogen protons is random (B)When an external magnetic field is applied the protons align with the field (C)A radio frequency (RF) pulse is applied which alters the spin of the protons and puts them in an elevated energy state (they absorb some of the RF energy) (D)When the RF pulse is turned off the protons release their energy (which is detected by the MRI machine) as they return to the 3. Combined measures of function and structure (fMRI, PET) Intro to fMRI here functional Magnetic Resonance Imaging (fMRI): Same process as MRI but this time the imaging is based on the ratio of oxygenated/deoxygenated hemoglobin. Deoxygenated hemoglobin is paramagnetic (weakly magnetic) whereas oxygenated hemoglobin is not Active regions receive large amounts of blood to supply oxygen to active neurons This value is called the Blood Oxygen Level- Dependent, or BOLD effect Indirect measure fMRI data The 3D image it provides is built up in units called voxels. Each one represents a tiny cube of brain tissue Each voxel can represent a million or so brain cells. The yellow blobs in the image below are actually clusters of voxels fMRI sources of noise: Noisy data! Moving subjects Things change over course of experiment External sources Pre-processing attempts to increase SNR fMRI pre-processing: Motion correction (“Don’t move!” – small movements lead to large signal changes)* Slice timing correction (slices acquired at different times)* Spatial filtering (smoothing across voxels) Temporal filtering (High/low pass) Registration/ Normalising (allows comparison across subjects by mapping to template brain)*  Co-Registration refers to the alignment and overlay of fMRI data from a single subject with that subject's own but separately acquired anatomic image (MRI) Normalization is a similar process that aligns and warps fMRI data from multiple subjects into a generic anatomic template. fMRI example experiment: Aim: Attentional bias to drug-related stimuli is predictive of outcomes in addicted individuals. To identify the neural substrates of this attentional bias which may help in identifying neural substrates critical to relapse vulnerability Method: 28 female nicotine dependent subjects Offline Smoking Emotional Stroop (SES) task – Neutral and smoking-related words displayed in red, green or blue – Participants were asked to identify the font colour as quickly and accurately as possible, using a button press, while ignoring word meaning. In scanner viewed smoking or neutral images Janes et al. (2010) Results: Whole-brain analysis correlating brain reactivity to smoking vs neutral images with SES task performance. A positive correlation was found between brain reactivity and SES interference effects in the insula, amygdala, hippocampus, parahippocampal gyrus, and occipital cortex. Top panel crosshairs are located in the right insula. Janes et al. (2010) Conclusion: The attentional bias toward smoking-related words was positively correlated with greater brain reactivity to smoking vs neutral images in brain regions involved in memory, emotion, and visual spatial processing. These findings suggest that smokers with elevated attentional biases to smoking-related stimuli may more readily shift attention away from other external stimuli and toward smoking stimuli- induced internal states and emotional memories. Such attentional shifts may contribute to increased interference by smoking cues, possibly increasing relapse vulnerability. Treatments capable of inhibiting shifts to drug cue-induced memories and internal states may lead to personalized tobacco dependence treatment for smokers Janes et al. (2010) Neuropsychopharm. Quick activity (1): 1. Write down three important details about EEG 2. Write down three important details about fMRI fMRI issues and considerations:  Good spatial resolution  Not invasive  MRI is less expensive than PET X MRI is way more expensive than EEG for example X Can cause claustrophobia X Chance of heating in subject (tattoos etc.) X Poor temporal resolution (sec) due to indirect measure X Very noisy Positron Emission Tomography (PET): Measures local variations in cerebral blood flow correlated with mental activity. Radioactive substance injected into blood stream (most common isotope is radioactive oxygen) Radiation from this “tracer” is monitored by scanner PET assumes that there will be increased blood flow to areas of high activity Images show the distribution of blood flow Positron emission tomography The patient is drip-fed a radioactive chemical (1) that works its way around the body. The scanner (2) emits low-energy gamma radiation (3) to the areas of the body under examination. The gamma radiation causes the radioactive chemical in the blood supply (4) to emit detectable photons (5) which are interpreted by computer software (6) to show how that part is functioning. Pet and addiction  PET brain scans show chemical differences in the brain between addicts and non-addicts. Addicts have fewer than average dopamine receptors in their brains, so that weaker dopamine signals are sent between cells (Volkow, 2001). How might this affect individuals (based on last week’s lecture)? PET issues and considerations: x Exposure to radiation x Because the radiation decays quickly PET can only be used for short tasks x Expensive  Different compounds can show blood flow, oxygen and glucose metabolism in the tissues of the working brain. These measurements reflect the amount of brain activity in the various regions of the brain  Diagnoses of wide range of diseases (a tracer such as glucose that collects in cells that are using a lot of energy, such as cancer cells). General limitations of fMRI (and neuroscience)  Subjects given explanations of psychological phenomena (some good, some bad) For half the explanations, the researchers added spurious neuroscientific detail, such as “brain scans indicated that...” or “because of the frontal lobe circuitry involved”. What they found was that people were far more likely to rate a bad explanation as satisfactory if it contained some neuroscientific material. Weisberg et al. 2008. J. Cognitive Neuroscience Sample sizes: 50 participants is considered big! Think about what this means for analysis… The effects we are looking for are likely smaller but our n has not increased. Larger studies have not always been encouraging: E.G. Thyeau et al. (2012) - 1326 subjects – Passive face viewing task – obtain a wide activation pattern, far from being limited to the reasonably expected brain areas, illustrating the difference between statistical significance and practical significance. Big (open source) data is the future… False positives (type 1 error) & false negatives (type 2 error) Neurons – non-direct measures Linking neuroscience and behaviour ? Human behaviour “with a few notable exceptions, the relationship between neural circuits and behavior has yet to be established” (Carandini, 2012) Does fMRI measure neuronal activity? fMRI measures blood flow in a region of neurons One voxel may contain 80,000 neurons so not specific Slow response when in fact neural communications are very fast (ms) How can we improve fMRI research? Larger sample sizes Open source code/analysis/data Publication of null results Pre-registration of studies All is not lost – many results have been replicated… Combine methods Use connectivity analyses Combined methods: Allow greater insight than a single method Often combine issues of both methods! However they should only be used if a single method could not do the same. Decision making Decisions require a representation of the rewards associated with choices Reinforcement learning (RL) suggests we use the prediction error (PE, the difference between expected and received rewards) as a learning signal to update reward expectations Prediction Error (PE) PE is the difference between the expected reward and the actual outcome (learning signal) Used to update future expectations and behaviour Wolfram Schultz first discovered that dopamine neurons code PE Striatum shows modulation of Prediction Error (PE) components PE & EEG ERP studies reveal early and late valence Single trial analysis - separate valence and magnitude PE components EEG - Good temporal (when) but poor spatial (where) resolution Philiastides et al. (2010 PE & fMRI literature fMRI data is more confusing, valence and magnitude representations often combined Striatum implicated in PE (along with some other areas) Receives projections from dopaminergic neurons. fMRI – Poor temporal (when) but good spatial (where) resolution Meta analysis technique 1.Formulate the problem 2.Literature search 3.Selection criteria (inclusion/exclusion) From Costafreda (2009) Meta analysis of PE literature Categorized fMRI studies of PE into three groups (based on EEG findings): 1. Valence (Positive/Negative) 2. Magnitude 3. Signed PE (includes VAL and MAG) Meta analysis results: (Fouragnan, Retzler & Philiastides – 2018. HBM) Separate networks for valence (positive and negative) and magnitude PE Only overlap in striatum Meta analysis results: (Fouragnan, Retzler & Philiastides – 2018. HB No unique signed PE representation Meta analysis summary Within valence PE, separate POS and NEG networks Valence and magnitude PE have distinct neural correlates Signed PE overlaps with other dimensions How do these networks interact and influence learning? Concurrent fMRI & EEG Meta findings cannot tell us anything about when things happen….but it does indicate that there are separate networks involved. Concurrent EEG/fMRI capitalizes on strengths of each technique – spatial ability of fMRI and temporal resolution of EEG Reversal learning task Fouragnan, Retzler, Mullinger & Philiastides (2015) Nat Com Concurrent fMRI & EEG – EEG results Replicate earlier EEG work… We use these single-trial values for the fMRI analysis Fouragnan, Retzler, Mullinger & Philiastides (2015) Nat Com Concurrent fMRI & EEG results Separate networks for early valence, late valence & magnitude Separate networks for early valence, late valence & magnitude Separate networks for early valence, late valence & magnitude Separate networks for early valence, late valence & magnitude EEG informed analysis reveals areas not identified in model based analyses (fMRI alone) Fouragnan, Retzler, Mullinger & Philiastides (2015) Nat. Com Concurrent fMRI & EEG summary Using EEG has revealed latent networks which would not have been revealed with conventional analyses We then continued this analysis and worked on the spatiotemporal profile of prediction error (see Pisauro, Fouragnan, Retzler & Philiastides, 2017. Nat. Com.) A better understanding may enable better treatments for disorders of decision making , e.g. addiction. Bringing it all together… Formulate your research question based on existing literature Choose your Build neuroscienc upon e method results carefully In groups start making a plan for the following based on today’s methods: Compare and contrast two neuroscience methods. Discuss the advantages and disadvantages of each with examples from the literature