[1] Hagmann, Patric et al. White matter maturation reshapes structural connectivity in the late developing human brain. PNAS 107:19067-19072, 2010.[2] Hermoye, Mori et al. Pediatric diffusion tensor imaging: normal database and observation of the white matter maturation in early childhood. Neuroimage 29:493-504, 2006.[3] Murphy, Shawn N. et al. High Throughput Tools to Access Images from Clinical Archives for Research. Journal of Digital Imaging, 2014.[4] Murphy, Shawn N. et al. Visual Interface Designed for Novice Users to find Research Patient Cohorts in a Large Biomedical Data-base. AMIA Annu Symp Proc: 489–493, 2006.[5] Doshi, Jimit et al. Multi-Atlas Skull-Stripping. Academic Radiology, Vol 20:1566-1576, 2013.[6] Ou, Yangming et al. Brain Extraction in Pediatric ADC Maps, towards Characterizing Neuro- Development in Multi-Platform and Multi-Institution Clinical Images. Under review. 2014.[7] Hüppi, Petra S et al. Diffusion tensor imaging of brain development. Seminars in Fetal and Neonatal Medicine 11: 489-497, 2006.[8] Dubois, J. et al. The early development of brain white matter: A review of imaging studies in fetuses, newborns and infants. Neu-roscience 276: 48- 71, 2014.[9] Lenroot Rhoshel K. et al. Brain development in children and adolescents: insights from anatomical magnetic resonance imaging. Neuroscience & Biobehavioral Reviews 30: 718-729, 2006.[10] Reiss, Allan L. et al. Brain development, gender and IQ in children: A volumetric imaging study. Brain 119: 1763-1774, 1996.

Clinically acquired Diffusion MR images were amassed and processed with our neonatal-specific automated brain extraction pipeline. Mean whole-brain ADC and volume values were extracted and used to characterize early life development.

This study demonstrates that clinical ADC images can be used to obtain mean-ingful quantitative measures of normative brain development in early childhood, confirming that clinical material is of sufficient quality to be used for research.

These are the critical initial steps towards the goals of mining and repurposing of institutional big data in order to further neurodevelopmental characterization.

We used the Research Patient Data Registry (RPDR) to identify a pediatric cohort and the mi2b2 workbench (mi2b2help.partners.org) [3] to retrieve clinical brain MRI scans.

RPDR (Figure 1) is an analytical database allowing researchers to access clinical information in order to obtain cohorts of patients with IRB approval [4]. It is the foundation upon which the nationally distributed i2b2 (Informatics for Integrat-ing Biology and the Bedside) project was developed (https://www.i2b2.org/).

Diffusion Tensor Imaging measures the water diffusion magnitude and is a surro-gate marker for myelin development [1], which undergoes dramatic changes in early neurodevelopment [2].

The aim of this study was to investigate an MRI metric, the Apparent Diffusion Co-efficient (ADC), of normative brain development from birth to early childhood.

We used the clinical informatics infracture at our institutions in order to identify this pediatric cohort and retrieve images from the Radiology Department archives. Clinical ADC images were then processed with multi-atlas-based skull-stripping tools and brain morphometric measures were extracted.

mi2b2 (Figure 2) is an open-source addition to RPDR/i2b2 that allows medical im-ages collected during routine clinical practice to be repurposed for research and educational use in a HIPAA-compliant manner [3].

There is inevitable risk of inaccuracies when manually reviewing medical records and visually inspecting images. A way to address this issue would be to increase our sample size. However, a rigorous statistical analysis is so far lacking to show how many subjects are sufficient to fully cover the inter‐subject variability and ef-fectively represent the normal range of variations.

Lastly, access to multi‐institution, multi‐vendor and multi‐platform data will help improve the representation power of our atlases. In the future we will test our framework’s performances in data from different vendors (GE, Phillips) and differ-ent magnetic field strengths (1.5T vs 3T).

Using clinical images to study the evolution of mean ADC values and brain volume of healthy pediatric subjects Kallirroi Retzepis1,2, Yangming Ou2, Lilla Zöllei2, Nathaniel Reynolds1,2, Victor Castro3, Steve Pieper5, Shawn N. Murphy3, 4, P. Ellen Grant6, Randy L. Gollub1,2

1Department of Psychiatry, Massachusetts General Hospital, Charlestown, MA, 2A.A. Martinos Center, Charlestown, MA, 3Information Systems, Partners HealthCare System, Inc, Charlestown, MA 4Laboratory of Computer Science, Massachusetts General Hospital, Boston, MA, 5Isomics, Boston, MA; 6Fetal-Neonatal Neuroimaging & Developmental Sci. Ctr., Children’s Hosp. Boston, Boston, MA

Methods

Results

Conclusion

Limitations

References

Mean ADC values display a dramatic decrease during the first six months of life (Figure 4) and begin to plateau starting at age one. This is consistent with pub-lished reports [7] [8] about rapid myelination during early infancy and its conse-quent stabilization during childhood.

Acknowledgments

This study was supported by NIBIB NIH RO1 EB014947.

In Figure 5, a rapid rise in average brain volume characterizes the first year of life, followed by a steady phase in early childhood. This is in agreement with published literature [9] [10].

Figure 2A screenshot of the first panel of the mi2b2 workbench - “Patients”.

In a typical workflow, users log into the client, and enter a list of patient-specific MRNs (Medical Record Number) or study-specific ANs (Accession Number). They can then locate the studies of interest [2] and initiate a request [3] to transfer them from the institutional repository to the mi2b2 cache [4]. Once this procedure is successfully completed the requested material can be downloaded to a local machine and then viewed and analyzed [5].

Figure 4ADC characterization of neurodevelopment.

Mean and standard deviation of whole-brain volume in each age group.

Figure 3A. Manual region of interest (ROI) segmentation in an example atlas (from year one to two)B. The change of mean ADC values with age.

In all plots the y axis is the mean ADC value (μm2/s). All ROIs were annotated by a trained expert in the series of con-structed ADC maps.

Figure 5Morphometric characterization of neurodevelopment.

Mean and standard deviation of whole-brain volume in each age group. CSF is excluded.

Introduction

Figure 1The main window of the RPDR query wizard. The user interacts with the panels in order to con-struct a query.

[1]: A hierarchical tree of items for users to choose from.[2-4]: A set of panels, each con-taining a number of items from [1]. The panels are related to each other via logical operators.[5]: When a query is run, aggre-gate numbers about the patient population matching the criteria are displayed.[6]: Users submit the query after obtaining IRB approval by click-ing on “Request Detailed Data“.

We submitted a request to RPDR for patients aged zero to six years old who had diffusion scans after 2000. It resulted in a cohort of 4745 pediatric patients with a brain MRI. 1765 potentially normative brain MRI studies belonging to 1600 pa-tients were identified. After expert review of the medical records to exclude any brain trauma or neuro-developmental abnormality and visual inspection of all images to exclude motion corrupted or artifact-degraded scans, we created a final cohort of n= 308 healthy pediatric subjects. This normative cohort was divided into 10 age groups sampling the first two weeks (neonatal stage) , quarterly in the first year, and then yearly af-terwards till six years old to most robustly capture developmental changes.ADC images were processed with multi-atlas-based skull-stripping tools specifi-cally implemented for neonatal and pediatric ADC images [5] [6] and brain mor-phometric measures were extracted (Figures 3, 4, 5).

Diffusion weighted images were acquired on a Siemens Trio 3T scanner at MGH. Repetition time (TR): 7500~9500 ms, echo time (TE): 80~115 ms, bvalue: 1000 s/mm2, image size: 256x256x60 voxels, voxel size: 0.86x0.86x2.0 mm3. ADC maps were gen-erated by scanner‐ embedded diffusion tensor imaging software.

1 2 3 4 5

1

2 3 4

5

corpus callosum

corpus callosum

Left Thalamus

RightThalamus

Left Caudate

RightCaudate

6

corpus callosum

A B