Rosenberg et al., A neuromarker of sustained attention from whole-brain functional connectivity

Question 1

What is a neuromarker?

Answer 1

A brain measure that is associated with a cognitive or behavioral outcome that can predict individual performance

Ie., a pattern of brain activity that reflects each person’s sustained attention ability

Question 2

What is functional connectivity?

Answer 2

Correlation in acrtivation between BOLD activity in different areas over time

• statistical dependence between time series of electro-physiological activity and (de)oxygenated blood levels in distinct regions of the brain

Question 3

What were 3 reasons/motivations for identifying a neuromarker of sustained attention, as stated by
Rosenberg et al.? (3)

Answer 3

behavioral measures are diverse
difficult to standardize (no summary index)
research is fragmented

Question 4

What are the attentional processes involved in sustained attention?

Answer 4

Information selection (& enhancement of selected information)

Inhibition of unselected information

Question 5

Describe the experimental task Rosenberg et al. used in the MRI scanner to capture these attentional
processes. Be sure to specify what aspect or component of the task was used to measure each attentional process. (2)

Answer 5

Participants watched images of cities and mountains and hit a button if the scene was a city. Because 90% of the scenes were cities, participants had to pay attention to make sure that they weren’t hitting the button by mistake, as they would be inclined to hit it every time. Attentional ability was measured by the number of mistakes made.
Information selection -
Inhibition of unselected information -

Question 6

What were the populations (groups of participants) they collected fMRI data from? For each population,
what were participants doing when fMRI data was collected?

Answer 6

Yale participants (25 students)
- performing gradCPT: d’ or sensitivity (a measure of accuracy that takes into account tendency to hit a button when you are in doubt; higher = better performance)
- fMRI data collected during gradCPT and resting state

Beijing participants (113 kids and teenagers, mean age 11; some with ADHD diagnoses and some controls)
- fMRI data collected during resting state only
(scores collected on ADHD-RS questionnaire)

Question 7

List two assumptions justifying/supporting Rosenberg et al.’s approach to analyzing the fMRI data and calculating the SAN model.

Answer 7

Individual differences in sustained attention will be reflected in complex patterns of correlated BOLD activity across brain regions

These patterns will be observed both when doing a task and at rest

Signature of these patterns should be able to predict attention ability in others

Question 8

What were the two neuromarkers of sustained attention they calculated for every individual?

Answer 8

Each individual’s positive and negative network scores (SAN).

• each strong edge was correlated with the participant’s good or bad performance on gradCPT
• stronger positive network score predicated (found) good performance on the gradCPT task
• stronger negative network score predicated poor performance on the gradCPT task

Question 9

What were the behavioural variables measured in each population? According to the authors, what was each behavioural variable an index (measure) of?

Answer 9

gradCPT: measures capacity for sustained attention
- lab measure of actual performance

ADHD-RS (ADHD Rating Scale): measures parent’s opinion of attentional abilities and ADHD symptoms
- questionnaire by another person about participant

Question 10

What were the three main relationships they reported between SAN neuromarkers and behavioural dependent variables? (3)

Answer 10

Sustained attention performance prediction by positive, negative, and both (general linear models)

• SAN neuromarkers: individual’s positive and negative network scores
• tail: network of nodes that fell out from their relationship with good and bad performance (tails reflect strength of all correlated activity between pairs of nodes whose correlated activity best predicted gradCPT performance); the two tails together made up <8% of the total number of edges

Question 11

What are SAN models?

Answer 11

Sustained Attention Network model: a brain network based statistical model (index of strength of connectivity)
- correlated activity across regions is reduced to two numbers that reflect degree of connectivity in a brain network associated with the capacity for sustained attention
- brain network scores are used to predict individual’s attention performance –> positive network strength predicts high attention; negative network strength predicts low attention

Question 12

According to the authors, what is the primary advantage of using the SAN model as a neuromarker? What is the advantage of using resting state data?

Answer 12

SAN models allow functional connectivity between many nodes from many networks to predict cognitive ability across different populations
- predictive, not descriptive
- data collection is relatively easy and quick

Resting state data
- easy to collect
- unbiased
- generalizable
- useful in populations with difficulty performing tasks (not confounded by differences in task performance)

Question 13

According to the authors, what are the two main implications of their findings for understanding of neural underpinnings of sustained attention?

Answer 13

models based on functional brain networks are powerful, generalizable predictors of cognitive abilities; can serve as a holistic neural index of sustained attention

attentional mechanisms extend beyond traditional attention regions and networks – coordinated activity across cortex, subcortical structures, and cerebellum

meaningful overlap between the neural mechanisms that are important for sustained attention and the neural dysfunction that leads to an ADHD diagnosis

Question 14

What did the authors study?

Answer 14

Patterns of large-scale network activation to predict
attentional abilities related to ADHD symptoms

Question 15

What was the big picture problem?

Answer 15

Attention is key for perception and cognition, but different types of attentional processes are measured in too many different ways; we need a summary index

Q: Can we find a neuromarker (brain-based measure) of general attentional ability?

Question 16

What is a summary index?

Answer 16

A single measure or number that sums up one person’s general ability

Question 17

Why might the authors want to look at functional connectivity across the WHOLE brain and not in, say, the dorsal attention network?

Answer 17

Sustained attention involves a variety of functions in a wide range of regions, cortical, subcortical, includes cerebellum

Question 18

What are intrinsic networks?

Answer 18

Correlated activity observed with fMRI - slow BOLD fluctuations between regions that go up and down in unison during rest or doing a task

Question 19

Why do Rosenberg et al. say that a measure of
attention based on resting state data would be a really useful thing?

Answer 19

Resting-state data is relatively straight forward to collect and share across different locations and across language and cultural barriers

Ie., can measure resting-state activity in populations who have different levels of ability to perform tasks in the scanner (can’t lie still/focus for extended periods)

Question 20

What was the specific question?

Answer 20

Can we take a data-driven approach to pulling out patterns of network activity to give us a marker of sustained attention that will generalize across populations?

Question 21

What is a Graduated Continuous Performance Task (gradCPT)?

Answer 21

• lie in the scanner and watch images of cities and mountains as one image dissolves into another
• press a button when you identify a scene as a city scene; do not respond if it is a mountain scene
• 90% of the scenes are cities –> pressing button is the default

Question 22

What is d’?

Answer 22

d prime: a measure of an individual’s ability to detect signals; sensitivity

• how accurate people are at hitting the button when they should, taking into account tendency to just hit the button all the time
• higher d’ means you are better at the task

Question 23

What is the ADHD-RS measure?

Answer 23

A questionnaire, filled out by parents; 18 questions
- inattention assessment (9)
- hyperactivity and impulsivity assessment (9)

Question 24

What were the variables?

Answer 24

IV
- gradCPT conditions (city vs mountain)
- choice of 268 network nodes
- participant groups (3 - Yale, Beijing ADHD, Beijing control)
- summary measures of network connectivity based on BOLD time courses (measures of positive and negative network strength that predicted good (+) and bad (-) attentional ability) –> used as predictor variables on both studies

DV
- BOLD response
- performance on gradCPT (d’)
- ADHD-RS score
- summary measures of network connectivity based on BOLD time courses (second-order DV) –> developed on Yale group

Question 25

What is a node?

Answer 25

268 spheres containing a few voxels; chosen based on previous research
- each node was part of the brain that played an important role in one of the canonical networks
- to find patterns of connectivity, define strength of connection between nodes
- edge: correlation between activity of two nodes

Question 26

How did they get the two numbers from the Sustained Attention Network model?

Answer 26

1. Correlated each edge (connection between two nodes) with performance on the sustained attention task (gradCPT)
2. Divided edges into two categories: edges whose strength was MOST associated with better performance / edges whose strength was associated with worse performance
3. Each network of nodes (correlated with better vs worse performance) was called a tail.
4. The correlations of all the edges in each tail summed up –> summary statistics of network strength
   - positive network strength predicts high attention
   - negative network strength predicts low attention

tail: tails reflect strength of all correlated activity between pairs of nodes whose correlated activity best predicted gradCPT performance; the two tails together made up <8% of the total number of edges

Question 27

How did they use the SAN model to predict scores?

Answer 27

Of the 25 Yale study participants, they removed data from each (one at a time) and used the summary scores from the rest of the group to predict the removed participant’s score on the gradCPT

They compared the predicted scores to the actual scores to make the scatterplot graphs (showing correlation between predicted d’ and actual d’)