COINS: An Innovative Informatics and Neuroimaging Tool Suite Built for Large Heterogeneous Datasets

Abstract

The availability of well-characterized neuroimaging data with large numbers of subjects, especially for clinical populations, is critical to advancing our understanding of the healthy and diseased brain. Such data enables questions to be answered in a much more generalizable manner and also has the potential to yield solutions derived from novel methods that were conceived after the original studies' implementation. Though there is currently growing interest in data sharing, the neuroimaging community has been struggling for years with how to best encourage sharing data across brain imaging studies. With the advent of studies that are much more consistent across sites (e.g., resting functional magnetic resonance imaging, diffusion tensor imaging, and structural imaging) the potential of pooling data across studies continues to gain momentum. At the mind research network, we have developed the collaborative informatics and neuroimaging suite (COINS; http://coins.mrn.org) to provide researchers with an information system based on an open-source model that includes web-based tools to manage studies, subjects, imaging, clinical data, and other assessments. The system currently hosts data from nine institutions, over 300 studies, over 14,000 subjects, and over 19,000 MRI, MEG, and EEG scan sessions in addition to more than 180,000 clinical assessments. In this paper we provide a description of COINS with comparison to a valuable and popular system known as XNAT. Although there are many similarities between COINS and other electronic data management systems, the differences that may concern researchers in the context of multi-site, multi-organizational data sharing environments with intuitive ease of use and PHI security are emphasized as important attributes.

Keywords: brain imaging; database; neuroinformatics.

Figure 1

Data sharing taxonomy. (A) e.g., A DICOM file may fit within the taxonomy as follows: plurality = singleton; medium = computer file; confidentiality = sensitive, anonymized; sense = data and metadata; source = recording of observation; mode = instrument. if the dicom file were encrypted then confidentiality = sensitive, encrypted, anonymized. (B) Sharing action. If a PI were to share subject data with another PI inside of COINS, the action would fit in the taxonomy as follows: source entity = PI, target entity = PI, sharing operation = intersect subjects, delivery venue = in situ, transfer method = computer network, and security = source encrypted, transfer encrypted, target encrypted.

Figure 2

Overview of COINS.

Figure 3

Based on a schedule of events, each subject’s progress is tracked, and a researcher can see the progress at-a-glance. URSI is the unique research study identifier that is unique across all studies; C, the green squares = complete; M, the red squares = missing; NC, the gray squares = not complete.

Figure 4

When the first and second entries of an assessment do not match, a coordinator can resolve differences using a simple web interface.

Figure 5

ASMT supports user-defined calculated fields, including conditional if statements and a built-in reverse scoring function, among others. In this example, the formula divides the question id, UPSA3_16 by 4. If you wanted to make this question id UPSA3_17 equal to UPSA3_16 reverse scored, the formula would be as simple as revscore ([UPSA3_16]).

Figure 6

Upper portion of query builder specifying the handedness field for a specific study.

Figure 7

Second portion of screen in query builder requesting all rst (rest) scans for one study to be combined with all handedness in a different study (Figure 6). This is similar to a union of all subjects between the two studies matching the specified criteria.

Figure 8

Visualization of how PHI is unlinked for one subject in Study z (lower right box). Recall that URSI is a unique code randomly generated for identifying a subject in a study. Note that CoPI Bovet, right two bottom boxes, is CoPI for study y and study z. PI Barany, middle box, cannot deduce PHI for unlinked URSI C using URSI B, but CoPI Bovet may be able to if he has a good memory or some document linking URSI C to URSI B since he has privilege in both studies and URSI B is still linked. Compare to PI Adrian and CoPI Axelrod in the lower left box: they are blinded to the other studies y and z, regardless of linking. Summarily, if the user is in at least two of the studies the subject is in, there is a potential exposure, and this is outside of the technology implementation.

Figure 9

(A) Quick look provides a full overview of each analysis stage, showing incomplete, complete, bad data, and notes from scanner and analysis input (B) based on user-specified QA thresholds, preprocessing status is viewable and associated files may be downloaded (option available behind zoomed image) (C) investigators may view fMRI statistics, on each task, for various contrasts.

Figure 10

Key-only, ER Diagram of a portion of MRN’s database: mrs_subject_types is used for tracking controls, withdrawn, excluded, patient, and other user-definable classes of a participant in a study.

Figure 11

SQL statements listing the study and subject with a “rest” task.

Figure 12

Comparison of SQL statements to retrieve all scans, all modalities.