---
title: Introduction to Neuroscience (INS) v3
tags: Talk
description: View the slide with "Slide Mode".

---

---

# Introduction to Functional Magnetic Resonance Imaging (fMRI)

###### Will Strawson, [e-mail]

----

### By the end of this class you should...

###### - Understand the difference between structural and functional MRI

###### - Understand the biological basis of the fMRI signal

###### - Understand the data structure of fMRI

###### - Understand what 'activation blobs' do and don't represent

###### - Have an appreciation for preprocessing fMRI data

###### - Be able to compare fMRI to other neuroimaging techniques

---

# Magnetic Resonance Imaging


----

##### Image of human finger from Mansfield and Maudsley (1977) obtained at 0.35 Tesla in 23 minutes.

----

##### First whole body image , obtained July, 1977. It required nearly 5 hours to produce.

----


#### T1 weighted structural MRI image, obtained ~2018. Took approximately 7 minutes to produce!
##### CSF signal repressed

---

# Structural MRI

#### Structural images provide information about the anatomy of what's being imaged by showing differences between different kinds of tissues (Grey Matter vs White Matter vs CSF)

#### Structural MRI provides high spatial resolution (0.8mm x 0.8mm x 0.8mm) images.

---

# Voxels


----

## Over to you (Viewing a structural image)
1. Open FSLeyes (using `terminal`)
2. `File -> Add from file -> fsl_course_data/ExBox11/structural.nii.gz`
3. Increase brightness + contrast

----


##### _Voxels_: Volumetric Pixel - the basic unit of measurement. Each voxel has a coordinate in the volume

##### _Slice_: all voxels in 2D plane.

##### _Volume_: The entire 3-D grid covering the space imaged in the MRI scanner. Volume is composed of voxels

##### _TR_: One volume scanned every TR - basic time measurement in MRI

---

# What about the 'f' in fMRI?

### We want to map _function_ to _structure_.

### But how do we quantify neuronal activation using non-invasive MRI?

---

## BOLD fMRI!
### _BOLD_ = Blood Oxygenation Level Dependent

#### We measure the level of blood oxygenation throughout different cortical regions, and infer neuronal activation from that.

##### Based on the premise that increased blood oxygenation = increased neuronal activation.

---

# Time

### Each voxel has an associated time-series of BOLD activation.

---



###### We have measurment of BOLD for each voxel across time

----

## Over to you (viewing a function MRI image)
1. Delete any previous images
2. `File -> Add from file -> fsl_course_data/ExBox11/fmri.nii.gz`
3. Expore volumes (movie view)
4. View time series' (View -> Time series)

---

# Preprocessing

### fMRI signal is _very_ noisy

### Many _preprocessing_ techniques are available to remove physiological and scanner noise e.g. temporal filtering ; Motion correction; spatial smoothing etc

---

## Brain Extraction / Skull stripping

1. `Add from file -> structural_brain.nii.gz`

---

## Coregistration

#### Co-Registration (or simply registration) refers to the alignment and overlay of fMRI data from a single subject with that subject's own but separately acquired anatomic image.

## Normalization

#### Process that aligns and warps fMRI data into a generic anatomic template (e.g. Talairach and MNI atlases). Normalization is usually performed on data from multiple subjects.

----

#### 1. Add standard `MNI152_T1_2mm_brain.nii.gz`

#### 2. Add `fmri.feat/reg/hires2standard.nii.gz`
- See how they match?

---

# Multiple Comparisons

### Question: with p-value = 0.05 for a single voxel and 100,000 voxels, how many false positives?

----

### 5000 of the 100,000 total voxels in a study (100,000 x 0.05) would potentially appear falsely activated.

### This issue is known in statistics as the multiple comparisons problem.

----

## Bonferoni Correction

### Bonferroni is super simple —just divide your original acceptance threshold (P≤0.05) by the number of tests you are analyzing.

###### e.g. 0.05 / 100,000(voxels) = p < 0.000001

#### However, this will equal high false negative results,

##### Other methods: use 'False Discovery Rate (FDR)'; Cluster Correction.

----

###### (Craig Bennet et al)

---

# What do those blobs mean?

### BOLD -> Parameter Estimate -> Test-statistic -> p-value


----

## Over to you (Overlaying statistical map)
1. `File -> Open from file -> fsl_course_data/ExBox11/fmri.feat/thresh_zstat5.nii`
2. Change colour grading from grey
3. Notice the scores on the bottom right

---

## Temporal and Spatial Resolution


###### - In comparison to other methods, fMRI has very good spatial resolution (~3mm) but average temporal resolution (~3s).
###### - One axis is missing from this diagram: Invasivness.

---

## One Caveat...

#### The fMRI methods of analysis I've presented, and the language i've used, may implicitly assume a modular view of the brian.

###### e.g. Different cognitive processes require different computational solutions which are implemented in anatomically/functionally distinct regions that operate independently – that is, in a modular fashion - brain area _x_ is responsible for cognitive process _y_

----

#### This is a BIG assumption that is still up for debate.

#### There is evidence that cognitive processes relying on highly distrubuted networks of coordinated activity, in specific frequency bands (see EEG/MEG evidence).

----

#### What do you think?

---

## Summary

###### - Structural-MRI = hi-res, no time; functional-MRI = has time componant

###### - Biological basis of fMRI = BOLD

###### - Data structure composed of voxels; slices; volumes. (4 dimensional for fMRI)

###### - 'blobs' represent statistical scores not 'activation' or even BOLD

###### - fMRI data needs preprocessing

###### - fMRI has good spatial resolution, but poor temporal resolution in comparison to other neuroimaging techniques (e.g. MEG )

---

#### A good, chilled out resource: