DRIVING: QUANTIFICATION AND DRIVING: QUANTIFICATION AND APPLICATIONS IN NEUROPSCHIATRYAPPLICATIONS IN NEUROPSCHIATRY
Godfrey Pearlson, M.D.Godfrey Pearlson, M.D.
Vince Calhoun PhD.Vince Calhoun PhD.
OVERVIEWOVERVIEW
This presentation consists of three parts:
– A general introduction to driving studies
– An fMRI study of simulated driving in sober and intoxicated subjects
– A validation of a driving simulator vs. on-road driving in an instrumented vehicle in sober and intoxicated subjects
PART 1: PART 1: WHY STUDY DRIVING? #1WHY STUDY DRIVING? #1
Driving is a behavior. Clinicians are frequently asked to judge the appropriateness of motor vehicle driving in patients with neuropsychiatric conditions (e.g. dementias, bipolar disorder).
Despite this, there is relatively little research on the development of quantitative measures for assessment of driving safety.
Vehicle driving consists of a complex series of quantifiable motor/cognitive behaviors, including divided attention, perception, planning visuo-motor integration, vigilance, tracking, working memory, psychomotor control and judgment.
These behaviors are affected by aging, some prescription medicines, neuropsychiatric illnesses and substance use.
WHY STUDY DRIVING? #2WHY STUDY DRIVING? #2
Altered driving behaviors have important public health consequences.
For example, in the U.S. more than 3 million persons were reported injured and over 40,000 persons died in motor vehicle crashes in 1996.
Traffic accidents are the greatest single cause of death in 5-32 year olds.
Most collisions are due to human performance problems. Many are due to intoxicated drivers.
AUTOMOBILE DRIVING IS A AUTOMOBILE DRIVING IS A MULTI-TASK COGNITIVE MULTI-TASK COGNITIVE
ACTIVITYACTIVITY
Continuous Tracking (e.g. keep in lane)Vigilance (Awareness of other vehicles,
pedestrians)Divided Visual Attention (pay attention to
simultaneous events in different places)Perceptual Judgment (how close to wall)Memory (e.g. what’s seen in mirror)
Driving
Working MemoryDivided Visual AttentionVisuo-motor Integration
Visual Reaction Time
Simple Visual Perception
“Top Down”Emergent Properties
“Bottom Up”Specific Components
ASPECTS OF DRIVER BEHAVIORASPECTS OF DRIVER BEHAVIOR
1. Performance -related - e.g. perception, attention
2. Motivational - e.g. sensation-seeking, aggression
3. State variables and - e.g. age, mood, fatigue and individual differences intoxication
Obviously, the three levels relate in complex ways,in such behaviors as speeding
PROBLEMS WITH REAL ON-ROAD DRIVINGPROBLEMS WITH REAL ON-ROAD DRIVING
• Difficulty of obtaining quantitative measures
• Potentially dangerous
• Must be constrained for safety - hence not veridical
• Cannot set up conditions of most interest (e.g. pedestrians, near misses with other vehicles, etc.)
ADVANTAGES OF SIMULATED DRIVINGADVANTAGES OF SIMULATED DRIVING
• Safety
• Repeatability of measures
• Interaction of driver and environment
• Ease of obtaining quantitative measures
• Can simulate any condition of interest
PROBLEMS WITH SIMULATED DRIVINGPROBLEMS WITH SIMULATED DRIVING
• Generalizability / Validity compared to on-road driving?
• Subjective realism poor except in very expensive setups
• Simulator sickness (vestibular) in absence of motion base - especially in women
• Too “game-like” – need contingencies (e.g. fines)
• Immersive / VR environments needed
Sociology “Road Rage”
Brain Diseases Schizophrenia
Parkinson’s
Huntington’s
Stroke
AIDS Dementia
Seizure disorders
Pharmacology Prescribed drugs
(e.g. BZ, Neuroleptics)
caffeine, alcohol, MJ
Part 2: Validation Study of Part 2: Validation Study of Computer Simulated Driving Computer Simulated Driving
with Alcoholwith Alcohol
BACKGROUNDBACKGROUND
Computerized driving simulators are one of the most common tools used in driving research
It is unclear whether simulated and on-road driving are truly comparable; this is especially true for low cost, fixed base driving simulator systems
In particular, no study has directly addressed the issue of driving simulator validity in studies of ethanol intoxication
Study OverviewStudy Overview
10 subjects completed a driving task both while sober and under influence of alcohol in two experimental setups (modes): a driving simulator and instrumented vehicle on a specialized road
Ethanol and realistic placebo drink were administered in randomized, single blind fashion
Directly comparable measures of driving performance were collected from the instrumented vehicle and driving simulator
Subject blood alcohol content (BAC) and subjective intoxication ratings were measured throughout experiment.
MaterialsMaterials
On-road driving facilities: Virginia Tech Smart Road, a 1.7 mile closed circuit two lane highway
Instrumented vehicle: ’97 Olds. Aurora (automatic transmission) with sensors, accelerometers, and computerized data collection and storage
Simulator: STISIM Drive 100 model, fixed base with steering wheel, foot pedals, and high quality computer monitor output.
Data collection: Vehicle and simulator share many output variables (velocity, turning rate, acceleration, etc.) sampled at 10Hz
Virginia Tech Smart RoadVirginia Tech Smart Road
STISIM Drive 100 SimulatorSTISIM Drive 100 Simulator
A subject seated at the A subject seated at the simulator simulator
The simulator outputThe simulator output
Design ConsiderationsDesign Considerations
STISIM course designed to faithfully replicate the geometry and features of the Smart Road
~7 minutes each on STISIM and Smart Road minimized time-on-task effects; 25mph speed limit on both road and simulator minimized kinesthetic feedback differences between the two
Ethanol dosing individualized to produced consistent BAC across subjects (0.07±0.015%)
Ethanol and placebo administration randomized, and the placebo masked with small amount of ethanol to minimize expectation effects
AnalysisAnalysis STISIM and instrumented vehicle shared many output
variables; we analyzed the intoxication effect within each mode separately and then directly compared the magnitude of the two effects.
Some measures of driving performance were not identical between modes, but were similar; we indirectly compared these output variables.
Some variability in subject BAC’s was present; we used BAC as a continuous rather then discrete predictor variable.
ResultsResults Table 1 shows two measures of longitudinal
vehicle control that are directly comparable between the simulator and on-road modes: time spend over the speed limit and the summed change in speed over the course of the entire experiment.
Tables 2 and 3 show measures of latitudinal vehicle control that are similar, but not directly comparable between modes: lateral range as reported in the simulator and the number of times subjects were verbally reminded (by a passenger side observer) to stay within their lane during the on road driving course
ResultsResultsChange in outcome t-Test >0
per unit BAC p
Time over SIM 596.6 ±1386.4 0.116speed limit CAR 452.9 ±596.6 0.026(sec) Difference 143.7 ±1558.5 0.789
Summed SIM 540.8 ±955.1 0.064change in CAR 183.3 ±478.0 0.142speed (m/s) Difference 357.5 ±1221.9 0.402
TABLE 1: Longitudinal vehicle controlTABLE 1: Longitudinal vehicle control
ResultsResults
Change in outcome t-Test > 0
Lateral per unit BAC p
Range (m) SIM 2.63 ± 2.86 0.02
Difference
Lane Placebo minus EtOH p
Reminders CAR 1.13 ±1.50 0.05
TABLES 2 and 3: Latitudinal vehicle control in both TABLES 2 and 3: Latitudinal vehicle control in both modesmodes
ConclusionsConclusions
Specific measures of latitudinal and longitudinal vehicle control (weaving and speeding) are similarly sensitive to ethanol intoxication effects in both the simulator and real road task.
There is good validity for time over speed limit, summed change in speed and lateral range variables on our fixed base simulator as compared to on-road driving in this paradigm
A comprehensive description of the study is in: McGinty et al. 2001; Assessment of intoxicated driving with a simulator: A validation study with on road driving
DWI-fMRI PERSONNELDWI-fMRI PERSONNEL
Vince CalhounVince McGintyTodd WatsonIllyas SheikhRegina ShihGeorge RebokGeorge BigelowSteven YantisDavid ScottDavid AltschulSusan CourtneyGodfrey Pearlson
SIMULATED DRIVING: SIMULATED DRIVING: Part 3:Part 3:
QUANTIFICATION, VALIDATION AND QUANTIFICATION, VALIDATION AND fMRI STUDIES OF NORMAL DRIVING & fMRI STUDIES OF NORMAL DRIVING &
DRIVING WHILE INTOXICATEDDRIVING WHILE INTOXICATED
How To Analyze fMRI StudiesHow To Analyze fMRI Studies
The Scanner EnvironmentThe Scanner Environment
Detection/Detection/EstimationEstimation
fMRI process chain
RegistrationRegistrationFunctional ImagesFunctional Images
ThresholThreshold/d/
OverlayOverlay
Phase FixPhase Fix
TimeTime 11 22 33 ……750 750 (secs)(secs)
112233
0s0s.66s.66s.33.33ss 11 22
y Xβ e
NormalizationNormalization
11 22
DATA DRIVEN APPROACH (ICA)DATA DRIVEN APPROACH (ICA)
Overview of Process:– Work with entire data set at once (not just one voxel)– The algorithm separates the data into spatially &/or
temporally independent components (1 map and 1 time course for each component)
Advantage: flexibility, does not assume particular time course (or HR) for data set, different sources represent different functional domains
Disadvantage: results must be monitored carefully to ensure the data is being properly characterized
Voxels
Tim
e
Data(X) = Components (C)*1ˆ W
Time courses
Spatially Independent Components
MixingMatrix
Independent Component AnalysisIndependent Component Analysis
Voxels
Tim
e
Data(X) = *G
“Activation maps”
Corresponding to columns of G
β
Time coursesDesignMatrix
General Linear ModelGeneral Linear Model
The GLM is by The GLM is by far the most far the most
common common approach to approach to
analyzing fMRI analyzing fMRI data. To use data. To use this approach, this approach, one needs a one needs a model for the model for the
fMRI time fMRI time coursecourse
In spatial ICA, In spatial ICA, there is no there is no
model for the model for the fMRI time fMRI time
course, this is course, this is estimated along estimated along
with the with the hemodynamic hemodynamic
source locationssource locations
General Linear Model
1. Model1. Model(1 or more(1 or moreRegressors)Regressors)
oror
RegressionRegressionResultsResults
2. Data2. Data
3. Fitting 3. Fitting the Model the Model to the Data to the Data at each at each voxelvoxel
ix j
y j
01
ˆ ˆM
i ii
y j x j e j
The ICA model assumes The ICA model assumes the fMRI data, the fMRI data, xx, is a , is a
linear mixture of linear mixture of statistically statistically
independent sources, independent sources, ss..
Independent Component AnalysisIndependent Component Analysis
**
**
++
Goal of ICA is to Goal of ICA is to separate the separate the
sourcessourcesGiven the mixed Given the mixed
datadata
Source 1Source 1
Source 2Source 2
A 1 2
Ts ss
fMRI data, fMRI data, xx
1ˆ i is A x
1 ,...,T
Ni x i x i x
1 ,...,T
Ni s i s i s
wherewhere
11
,...,N
N ii
p s s p s
1s v
Data Generation(synthesis)
(b) Data Reduction
Data Processing(analysis)
(c) ICA
1
N
1ˆ A .T
(a) Preprocessing, Normalization
Brain MR Scanner
Ns v 2s v
t1u v
tmu v
t2u v
B 2y i
Ky i
1y j
2y j
Ky j
1x j
Nx j
2x j1ˆ B
1s j
ˆNs j
2s j1ˆ AA
1y i
Model for Applying ICA to fMRIModel for Applying ICA to fMRI
Multi-Subject ModelMulti-Subject Model
METHODSMETHODS
20 Subjects/50 scans Scan Parameters
– Single-shot EPI– FOV = 24cm, 64x64– TR=1s, TE=40ms– 18 slices– Slice thickness = 5mm– Gap = .5mm
Procedure– Subjects were trained to asymptote
performance on driving simulator with a simulated driving game, ‘Need for Speed II’ (NFS II)
– fMRI Scan performed during driving paradigm
– Drug Administered (oral Marinol or ETOH or placebo)
– 2nd fMRI scan performed at maximal blood levels
Driving fMRI ParadigmDriving fMRI Paradigm
* Drive Watch
0 600
60
• The order of the watch/drive epochs was alternated across runs
• Subjects were instructed to:• Remain within 100-140
KPH (if successful received bonus)
• Stay in right lane• Avoid collisions
NFS IINFS II
•We show a QuickTime movie of a 23 year-old male subject, a non-user of recreational drugs. The movie shows a brief segment of simulated driving performance while intoxicated.
• The subject had practiced to asymptote on the driving simulation program Need for Speed II (NFS II) which was used as the in scanner active task.
• Movie shows a brief NFS II segment illustrating lane deviation (weaving), followed by a vehicle collision. At this time the subject’s self-rated impairment was 2 on a zero (least) to five (most) analog scale.
Run 1 Run 2 Run 3 Run 4
* * * *
P/D D/P D/P P/D
* * * *
P/D D/P D/P P/D
* * * * P/D D/P D/P P/D
* * * *
<- 3 Min -> <- 2 Hrs -> <- 3 Min ->
• One of 4 variants is shown above (AB/AB; BA/BA; AB/BA; BA/AB)• Each epoch is 1 minute• Key: * = asterisk viewing P = passive viewing of driving D = active driving
Paradigm for the entire experimental session
*
P
D
*
P
D D
P
*
10 Min
1 Min
Experimental paradigm is a hemi-castle design
*
METHODSMETHODS
20 Subjects/50 scans Scan Parameters
– Single-shot EPI– FOV = 24cm, 64x64– TR=1s, TE=40ms– 18 slices– Slice thickness = 5mm– Gap = .5mm
Preprocessing– Timing correction– Motion correction– Normalization– Smoothing (6mm)
ICA– Data were reduced from 600 to 30
time points using PCA– Data from all subjects were
concatenated and further reduced to 25 time points
– Data were then entered into an ICA estimation utilizing the infomax algorithm
Neural Substrates of Simulated DrivingNeural Substrates of Simulated Driving
VD Calhoun, JJ Pekar, VB McGinty, T Adali, TD Watson, & GD Pearlson. “Different Activation Dynamics in Multiple Neural Systems During Simulated Driving Revealed by ICA of fMRI Data.” Human Brain Mapping 16(3), 2002.
Higher Order Visual/Motor: Increases during driving; less during watching.Low Order Visual: Increases during driving; less during watching.Motor control: Increases only during driving.Vigilance: Decreases only during driving; amount proportional to speed.Error Monitoring and Inhibition:Decreases only during driving; rate proportional to speed.Visual Monitoring: Increases during epoch transitions.
Color Regions Hypothesized Function
Green Bilat. cuneus, precuneuslingual gyrus
Visual monitoring
Yellow Cerebellum, inferior occipital
Low-order visual
White Bilat. visual associationBilat. parietal
Visuomotor integrn. High-order visuomotor
Red Cerebellum and motor cortex
Motor control
Pink Orbitofrontal and anterior cingulate
Error monitoring,Inhibition
Blue Medial frontal, parietal, post. cingulate
Vigilance
Interpretation of ResultsInterpretation of Results
Driving While Driving While IntoxicatedIntoxicated
DRUG EFFECTSDRUG EFFECTS
Goal 1: To visualize the neural substrate for operating a driving simulator as assessed by fMRI
Goal 2: To visualize the effect that alcohol or oral Marinol (THC) has on the neural substrate for operating a driving simulator as assessed by fMRI
Goal 3: To study the effects of these drugs on driving performance as assessed by a driving simulator
WHY STUDY MARINOL (SYNTHETIC THC)?WHY STUDY MARINOL (SYNTHETIC THC)?
• Increased prescribing for cachexia (AIDS, cancer).
• Claim of no behavioral effects.
• Need to study the effects on driving, operating machinery.
• Are persons legally prescribed THC, DWI?
• Yields stable, long-lasting plasma THC levels – a useful “test-bed” for smoked MJ.
Alcohol and Driving Performance (N=10 x 2)Alcohol and Driving Performance (N=10 x 2)
Marinol and Driving Performance (N=10)Marinol and Driving Performance (N=10)
ICA Time Courses (ETOH)ICA Time Courses (ETOH)
The activation during driving of the fronto-parietal (blue) regions is most significantly affected during ETOH intoxication
The modulation of primary visual areas (yellow) between driving and watching is preserved (unlike for THC)
Driving and AlcoholDriving and Alcohol
The activation during driving of the fronto-parietal (blue) regions is the most significantly affected during ETOH intoxication
This difference is dose dependent and increases as ETOH dose increases
The largest difference occurs during the first portion of the driving epoch
ICA Time Courses (THC)ICA Time Courses (THC)
1. The modulation of primary visual areas (yellow) between driving and watching is reduced during THC intoxication
2. The activation during driving of the fronto-parietal regions (blue) is also disrupted during THC intoxication
3. The anterior cingulate/orbitofrontal regions (pink) are decreased in amplitude during THC intoxication
Results and ConclusionsResults and Conclusions Meaningful top-down approaches are feasible in
an fMRI environment for a complex behavior
Driving activates a distributed network of areas including cerebellum, prefrontal, frontal eye fields, primary and secondary visual areas.
Preliminary imaging results reveal that:– Intoxication appears to modulate the temporal patterns
in brain regions rather than turn on or off different brain regions
– Alcohol & Marinol affect these temporal patterns differently, as well as having different behavioral “footprints”
Preliminary ResultsPreliminary Results
Alcohol– Vigilance regions appear to be affected in a dose-dependent
manner (in particular, the early portion)– Visuomotor integrative regions do not appear to be affected by
alcohol
Marinol:– Vigilance regions are significantly disrupted during intoxication– Visuomotor integrative regions are differentially activated by
driving versus watching only when subjects are not intoxicated– Error monitoring/disinhibition regions are specifically
activated during driving only when subjects are not intoxicated