How can we use and interpret neuro-imaging in practice?

Question 1

main interpretation pitfalls

Answer 1

1. correlation vs. causation
2. reverse inference
3. individual differences

Question 2

correlation vs. causation

Answer 2

• fMRI/EEG/MEG: correlational methods
• brain activity in area X correlates with performance on task A
• we can’t conclude that area X is necessary for (causally related to) performance on task A
• -> 3rd factors can be unknown

Question 3

correlation vs. causation: how can we investigate causal role of area X?

Answer 3

use interference/stimulation method: TMS (larger and superficial areas of the brain), pharmacology: imperfect specificity

Question 4

reverse inference: concrete example

Answer 4

1. women in bikinis activate area Z (current study)
2. tools activate area Z (previous studies)
3. therefore women in bikinis are viewed as tools
   - -> logic only works if area Z is involved in only one mental process
   - -> this is rarely (never?) true

Question 5

What is forward inference?

Answer 5

=OK

• what brain activity is associated with a given task/cognitive process?
• if cognitive process X is engaged, then brain area Z is active

Question 6

What is reverse inference?

Answer 6

= not okay (unless cautiously used to raise novel hypotheses for future research)
- going backwards from a brain activation to a particular cognitive function

Question 7

reverse inference: abstract example

Answer 7

1. in our study, task A activated brain area Z
2. brain area Z was active in other studies for cognitive process X
3. Thus, the activity of brain area Z in our study demonstrates engagement of cognitive process A in task A

Question 8

Why can’t we say anything about individual differences?

Answer 8

• neuro-imaging findings of different brain function in a disorder: “group” results
• -> average of GROUP 1 different from average of GROUP 2
• brain anatomy and fMRI/EEG signals known to differ across subjects, as well as strategies to do the same task
• one individual activation pattern could be more similar to the average of the other group (e.g, 1,90m woman)
• -> individual interpretations/ diagnoses not possible!

Question 9

How can you say something about individual differences?

Answer 9

• more sophisticated analysis techniques: classifier algorithms
• never 100% accurate!

Question 10

What are design issues related to psychopathology?

Answer 10

1. sample choice: representative?
2. development: cross-sectional vs. longitudinal
3. diagnostic use not possible (individual differences)

Question 11

What are the problems with a small sample size?

Answer 11

1. results less reproducible
2. results less reliable
3. low statistical power: lower probability to detect an effect + detected effect has lower probability to be true (false positive)

Question 12

What are the problems with a sample of convenience?

Answer 12

• low generalizability to general population: e.g. ASD patients with high IQ, white (WEIRD) students

Question 13

What is the solution to the sample choice problem?

Answer 13

use larger and representative (epidemiological) samples

Question 14

What is a cross-sectional design?

Answer 14

• between-group comparisons, e.g., age groups, groups with/without disorder

Question 15

What is the problem with cross-sectional designs?

Answer 15

• can give incomplete or misleading results, as groups/ samples may differ in other aspects (e.g., older group: more severe subtype of illness)
  (in a lot of developmental disorders the symptoms decrease with age, so matching on symptom severity doesn’t really make sense)

Question 16

What is a longitudinal design?

Answer 16

• repeated measures (within-subjects)

- enables tracking of true developmental trajectories

Question 17

What is the problem with longitudinal designs?

Answer 17

1. time-consuming

2. drop-out

Question 18

DTI longitudinal study example

Answer 18

• measurement of fractional anisotropy in 2 areas important for reading in children when they begin reading and 2 years after
• -> the lower the initial connectivity (FA), the higher the reading scores 2 years later
• -> mature (myelinated) fiber bundles at the beginning –> connections less flexible to adapt for reading

Question 19

Diagnostic use in neurological disorders

Answer 19

MRI techniques can improve diagnostic classification: e.g., imaging markers as diagnostic criteria for Alzheimer’s disease

Question 20

Diagnostic use in psychiatric disorders/ psychopathology

Answer 20

• more subtle, more distributed neural systems involved

- huge individual variation

Question 21

What is a promising direction for using individual brain scans?

Answer 21

• pattern classification algorithms
• categorize based on individual pattern of brain activation
• e.g., Multi Voxel Brain Analysis (MVBA) –> pattern of voxels –> algorithm reads the pattern and classifies it into one or the other group

Question 22

What is sensitivity?

Answer 22

• % of affected people correctly classified as sick

Question 23

What does 90% sensitivity mean?

Answer 23

• 90% sensitivity: 10% of sick people classified as healthy

- -> false negatives

Question 24

What is specificity?

Answer 24

• % of healthy people correctly classified as being healthy

Question 25

What does 90% specificity mean?

Answer 25

• 90% specificity: 10% healthy people classified as sick –> false positives

Question 26

What does the value of sensitivity/specificity depend on?

Answer 26

the prevalence of a disorder!