Table 1.
Study design recommendations to minimize sources of variation in fMRI studies.
| Source | Description | Recommendation |
|---|---|---|
| MR related | ||
| Scanner | Machine characteristics and performance (i.e. changes in scanner or changes in software or hardware on the same scanner), scanner stability | Perform data quality measurements (e.g., signal to noise) with phantoms before each data collection |
| Acquisition method | Pulse sequence and imaging parameters | Investigators should schedule regular maintenance |
| Placement | Differential subject position in bore | Longitudinal studies and or different sites agree upon “range of values” that all scanners should adhere in order to standardize measurements |
| Subject related | ||
| Subject | Individual differences in physiology, responses and hormonal rhythms | Conduct fMRI at similar times to minimize fluctuations due to circadian rhythms |
| Sample | Cohort size and composition | Determine by task design/cognitive construct/sample characteristics |
| Intrinsic | Noise and other unaccounted variation | Noise due to intra-individual variability could be minimized by increasing measurement occasions |
| Task related | ||
| Longitudinal processing | Voxel registration across timepoints to one another and to same anatomical location | Standardized MRI acquisitions followed by standardized pre-processing of imaging data to minimize differences |
| Motion during scans | Differences in motion across timepoints | Surface-based registration/analysis may help to reduce the influence of changes in cortical thickness which occur with development |
| Practice effects | Subjects might get better at the task at subsequent visits | Have an alternate version of the task or use adaptive methods where task difficulty is matched to each subject |
| Block between session | Variation across responses to each task presentation (attention, arousal, caffeine, etc.); non-task related cognitive processes; changes in cognitive strategy over time; task comprehension, attention and arousal | Task comprehension is easily solved by practice sessions before going into the scanner |
| Task | Block vs. event-related designs; target region | Event-related designs generally need longer task designs; investigator should consider the best approach |
| Time between session/Lag time | Temporal artifacts (e.g. drift, low-frequency oscillations, etc.) | Times between sessions should be small enough that there is no “developmental change”. This would vary depending on the cognitive construct and would need to be trialed by the investigator |