| Abstrakt: |
Background: Large‐scale datasets are becoming the norm in estimating the brain‐behavior relationships, which are acquired via engagement with participants over several years often across multiple sites and scanners. However, maintenance of imaging protocols in MRI centers is still an ad‐hoc process. Acquisition parameters, including but not limited to flip angle (FA), repetition time (TR), and echo time (TE), are routinely improvised at individual sites to accommodate changes in software/hardware and to suit the needs of the patient. Non‐compliance in acquisition protocol can lead incompatible MR images hindering downstream analysis. For example, co‐registration with the structural counterpart becomes difficult if EPI is non‐compliant with the field map. Method: We present an open‐source tool mrQA to automatically identify variations in acquisition parameters. A dataset is parsed to extract the parameters for each run. and compared against a reference protocol. If unavailable, reference protocol is inferred based on the most frequent value for each parameter. We recommend using mrQA for datasets in DICOM format (even though mrQA supports NIfTI) to prevent any non‐compliant parameters at the MRI scanner itself because other formats such as JSON sidecar (for NIfTI) lack standardization and hence, are often incongruous with across datasets. Result: To assess the prevalence of protocol compliance, we evaluate over 20 public datasets on OpenNeuro (BIDS) and the ABCD dataset (FastTrack ‐ DICOM). The exploration is not meant to be a finger‐pointing exercise but to discover common pitfalls in MR acquisition that can confound the research we do. Table 1 and Table 2 summarize the results for OpenNeuro and ABCD datasets, respectively. The results are reported separately for each vendor as even the same parameters differ in the unit of measurement across vendors. Conclusion: We demonstrate the pervasive problem of protocol non‐compliance based on the analyses of many open datasets from OpenNeuro and the ABCD dataset. Our tool, mrQA can summarize acquisition parameters from a MR dataset, in order to discover any issues of non‐compliance. Even though, the effect of non‐compliance on signal intensity will depend on the specific tissue in context, explicit characterization should precede any analyses in case there is considerable non‐compliance in acquisition parameters. [ABSTRACT FROM AUTHOR] |
