Mbirn: MAD: Updates

From NAMIC Wiki
(Redirected from MAD: Updates)
Jump to: navigation, search
Home < MAD: Updates

2006.03.28 (C. Fennema-Notestine)

  • This project tested whether these data, which have been collected using best available methods for each site, could be meaningfully reanalyzed as a larger combined dataset by using rigorous data curation, image analysis, and statistical modeling methods.
  • We explicitly examined the hypotheses that hippocampal volume, as measured by the Freesurfer segmentation algorithm, would show the expected age-related volume decrease and left/right asymmetries using data collected for studies of AD and normal aging at four mBIRN sites (UCSD, WashU and MGH/BWH combined).
  • We demonstrated that legacy MR data from multiple sites can be pooled to investigate questions of scientific interest with appropriate control of methods, cohort, and statistical modeling:
    • replicated the expected age-related decline in hippocampal volume and hippocampal asymmetries within the pooled legacy dataset
    • demonstrated that the mixed effect model was the most powerful statistical model
  • The results have been shared with the scientific community at the Society for Neurosciences 2005 Feasibility of multi-site clinical structural neuroimaging studies of legacy data: Aging & Alzheimer's disease, and a manuscript is being submitted.
  • Of note, this study has been a key prototype for the development of the infrastructure for 1) data sharing across sites that followed federal HIPAA guidelines for de-identification (Defacing, BIRNDUP code), 2) common database schema for clinical and derived morphometry data at each site and a Mediator to bring the multi-site information together seamlessly (HID), 3) shared locations for data storage and retrieval (Storage Resource Broker), and 4) mediated queries that interrogate data contained in databases located at the two sites (HID Query Interface, 3Dslicer, Query Atlas).
  • Future efforts will examine the classification of AD and elderly control individuals within this legacy data and incorporate data from prospective studies.

2005.11.18 (C. Fennema-Notestine)

  • An overview of the MAD project and its goals was presented at the BIRN AHM meeting, and this is viewable by selecting "MAD presentation" at http://www.na-mic.org/Wiki/index.php/Morphometry_BIRN_Agenda. This work is summarized below.
  • Study 1 – Morphometric measurements in Healthy Elderly Controls
    • Added collaboration with Anthony Gamst (UCSD/fBIRN biostatistician) to improve statistical modeling of the multi-site data.
    • Presented poster at Society for Neuroscience 2005 conference in Washington D.C. on modeling the relationship between hippocampal volume normal aging across multiple sites ( Feasibility of multi-site clinical structural neuroimaging studies of legacy data: Aging & Alzheimer's disease). The results of the study suggest that:
      • legacy MR data from multiple sites can be pooled to investigate questions of scientific interest with appropriate control of methods, cohort, and statistical modeling.
      • the mixed-effects model, employing site as a random effect, best fit the data, accounting for site-specific effects while taking advantage of expected comparability of age-related effects across sites.
    • Additional work within the BIRN continues to refine the segmentation algorithm (BF) which may further decrease site-specific variance due to differences in acquisition sequence.
    • This work describing the methodology and feasibility of pooling legacy MR data across sites will be completed and submitted in January 2006 (CFN).
  • Study 2 – Clinico-anatomic relationships
    • Clinical correlation studies continue with hippocampal volume and memory measures across MGH and UCSD sites; these two sites have similar neuropsychological testing variables.
    • These explorations will be complete by the mBIRN AHM 2006, and may lead to a separate brain-behavior manuscript (BCD).
  • Study 3 – Diagnostic classification
    • The goal of this project is to determine the classification of controls and AD individuals using MR morphometric and neuropsychological data across the two sites with AD data (UCSD and WashU).
    • Linear and quadratic discriminant analysis is being applied.
    • This project will be completed and submitted for publication by the mBIRN AHM 2006 (CR).
  • UCI MCI data currently being processed by UCI and MGH with the primary goal of providing training in image processing tools.
    • 25 UCI skull stripped MCI datasets were uploaded to MGH
    • MGH performed subcortical segmentation on these data using the snapshot MAD FreeSurfer version with WashU atlas
    • Datasets requiring additional editing were returned to UCI
    • Edited data will be returned to MGH for review and re-segmentation where viable
    • During this process, the UCI team will receive supervised training in the advanced use of the Freesurfer software
    • For UCI’s larger datasets, this training will extend to the new FreeSurfer platforms to be released.
  • These data will be used to create a legacy repository so that the data will continue to be available for in-house BIRN infrastructure and tool development.
  • Morphing MAD…
    • Prior to the mBIRN AHM 2006 meeting, we will draft new goals for the MAD testbed, which will focus our work on current clinical collaborations within the mBIRN (ADNI and VETSA).

2005.09.20 (C. Fennema-Notestine)

  • MAD Goals: Pooling and analyzing image data across sites
    • Increase statistical power for studying relatively rare populations or subtle neuroanatomic changes
    • Provide benchmark data sets for the use of image analysis tool developers for testing and comparison validation
    • These studies aim to test the hypothesis that legacy collections of clinical and structural MRI data from different sites can be meaningfully reanalyzed as a larger combined data set by using rigorous data curation and image analysis methods.
  • Study 1 – Morphometric measurements
    • The primary aim herein is to replicate previous clinical neuroimaging work in the study of normal aging with the combined legacy datasets; explicitly test the hypotheses that hippocampal volume will show the expected age related volume decrease.
    • Data from 3 sites (UCSD, MGH/BWH, WashU) were used to examine the feasibility of combining neuroimaging data collected at different sites.
    • To reduce sources of variability, the data has been curated and analyzed with the same morphometric methods (subcortical segmentation/MGH).
    • Site effects were examined in healthy control samples. Preliminary analyses suggest the importance of controlling for gender or estimated cranial vault across sites, and define potential limitations related to differences in pulse sequence.
    • To assess validity of combining data across sites, the relationship between hippocampal volume and normal aging has been examined in healthy elderly participants at all three sites (SFN 2005).
    • This work is being incorporated into a manuscript related to the methods of combining and curating this data across multiple sites (CFN, RG, BCD).
  • Study 2 – Clinico-anatomic relationships
    • Clinical correlations continue with hippocampal volume and memory measures across MGH (BCD) and UCSD (CFN) sites. This will lead to a separate brain-behavior manuscript (BCD, RG, CFN)
  • Study 3 – Diagnostic classification
    • Pooling data from two MAD cohorts with both healthy elderly controls and participants with Alzheimer’s disease (UCSD and WashU), work continues to improve diagnostic classification of AD. The study aims to determine participant classification with automatically generated MR measures of hippocampus and ventricles, in combination with basic mini-mental state exam scores. This work has potential generalizability for other on-going multi-site neuroimaging projects in clinical studies of AD. This study is a MAD collaboration with Cooper Roddey and Anders Dale.
  • UCI MCI data currently being processed by UCI and MGH for future inclusion in MAD studies.

2005.07.26 (C. Fennema-Notestine)

  • Weekly MAD meetings continue to move analyses and manuscripts forward (C. Fennema-Notestine, R. Gollub, B. Dickerson, J. Turner, C. Roddey). Study plans are being structured into specific manuscript goals; documents are in progress to provide greater manuscript/authorship details (for Xythos). Below is an overview of study plans.
  • Study 1 Update:
    • NC hippocampal volume and normal aging accepted for presentation at Society for Neuroscience (Monday, Nov. 14, 2005; Program No. 410.5). This work will be incorporated into a manuscript related to the methods of combining and curating this data across multiple sites (CFN, RG, BCD).
    • ETIV correlations with manual methods complete, suggest good validation, details forthcoming. Any valuable results to be included in above manuscript.
    • Following up on site effects (esp. edges near fluid) that may be related to pulse sequence differences at WashU (MPRAGE) and UCSD (SPGR).
  • Study 2 Update:
    • Preliminary work begun on clinical correlations with hippocampal volume across MGH (BCD) and UCSD (CFN) sites, focused on CVLT data available from both sites. This will lead to a separate brain-behavior manuscript (BCD, RG, CFN)
    • Moving forward on group comparisons (NC vs. MCI vs. AD)
  • Study 3 Introduction and Update:
    • This study aims to use a discriminant analysis method to differentiate NC from AD scans across UCSD and WASH U sites (only sites with AD) using the Freesurfer volumetric results. This study is in collaboration with Cooper Roddey and Anders Dale with potential generalizability for the ADNI project.

2005.07.19 (J. Jovicich)
OUTSTANDING ISSUES REGARDING THE UCI DATASETS

  • Define and fix de-facing differences in Thang's (UCI) and Amanda's (UCSD) processing streams
  • Define MAD pooling plans for the 40-brains UCI - MCI dataset. Start their subcortical segmentation (atlas?)


UPDATES ON DATA PROCESSING ISSUES WITH MCI DATA FROM UCI

  • After some iterations, Thang sent to MGH the defaced 1023-subject to ensure that: a) the defacing was working fine, and b) that the right/left orientation was well defined using as reference the same subject's year4 dicom.
  • The plan was:
    • if those two issues were checked on subject 1023, we'd then proceed to check 17 UCI brains from the MCI cohort for which there were baseline and Year4 dicom scans.
    • if these two issues were checked on these 17 brains, then we'd use the method on the 40 UCI brains from the MCI cohort identified for pooling in MAD. Thang would visually check that the defacing didn't chop-off brain before sending to MGH.
    • after that steps above MGH would start the subcortical segmentation on the 40 UCI-MCI datasets identified for MAD.
  • The defacing that Thang used chopped-off brain in subject 1023.
  • When Amanda run the de-facing code on the same subject she got good results. Thus, Thang and Amanda are doing slightly different things and they have to agree on a meeting time to figure out what exactly is going wrong on Thang's steps.
  • I don't think I ever received the 40 UCI-MCI MAD datasets that Jess would like to have segmented, header-deidentified but with face information. So there is no UCI-MCI segmentation in the queue on MGH's end. My understanding has always been that until the defacing & right/left issue is solved this wouldn't start.
  • If Jess can share the 40 UCI-MAD data having only de-identified the image headers and not the face, then:
    • The MAD group should review the clinical information of these 40 UCI datasets to assess their value in pooling with the other MAD datasets
    • Thang should make sure that with this new pre-processing stream we get the the right/left orientation correct. To do this, he could use the 17 datasets that have the Year4 dicom.



2005.06.09 (J. Jovicich, T. Nguyen)
UPDATES ON DATA PROCESSING ISSUES WITH MCI DATA FROM UCI

  • The original 3D structural MCI baseline images do not have left/right orientation information. They only have anterior/posterior and superior/inferior orientation information.
  • We tried to work with the images but they would give an error when converting for tkmedit viewing. An option was to convert the original files to MGH format. This also helps decrease the number of files. However, we couldn't convert them using mri_convert, so we used a program Tosa wrote for us called uci_convert.
  • As of now, the only baseline subjects for which we have orientation information are the UCI subjects. The reason we have this information is because we have the raw data for Year4 scans.
  • So to establish the left/right orientation of the baseline scans we have compared each subject's Baseline scans to their Year4 scans, finding landmark features such as lesions and structural similarities. In this way Tosa's code can be checked to see it's converting to the correct image orientation. We seem to have corrected our problems and scans from both years correspond with each other.
  • A separate problem has been the anonymization of images. We anonimized these subjects using the mri_deface program from the BIRN-DUP pipeline. The script that Thang used was just a short in-house script that saves on the manual work. The output from this program appends the filenames with .strip.*, depending on the file.
  • Processing steps at UCI:
    • Start from 3D baseline images
    • Convert to MGH format using Tosa's program
    • Take MGH format and de-identify (output in analyze format)
    • Optional: reconvert the de-identified image to mgh format


2005.05.24 (C. Fennema-Notestine)

    • RB's paper finds .93 correlation across all groups for ASF and manual - this was not broken down by group, although reportedly similar.
    • BD's MGH work on T2-based manual ICV correlated at .89 with BF's eTIV
    • Need correlation between BD's ASF/BD's manual and BF's eTIV for WashU subset
    • CFN will do local UCSD correlation for datasets overlapping with separate study that has ICV
    • Relationship of volumes with age
      • eTIV not related to age
      • For brain volume, r increased with use of proportionalized brain volume to eTIV
      • For hippocampal volume, r increased or stayed the same when prop. to eTIV. Also noted that r decreased when proportionalized to brain volume rather than to eTIV, suggesting eTIV is better solution.
    • Group volume comparisons and related effect size (partial eta squared) for NC vs. MCI vs. AD and just NC vs. AD
      • Effect size of Group increases (and F and level of sig "improve") when eTIV brought in as a covariate for both brain volume and hippocampal volume measures. This is true across and within each site.
      • Effect size same or less when brain volume is covaried rather than eTIV.

2005.05.17 (C. Fennema-Notestine)

    • MAD t-con Update and Progress on Oct 2005 Action plans:
      • Completed data extraction with subcortical segmentation (B Fischl), same version and atlas.
      • Analysis of cross site control group data (UCSD, MGH/BWH, WashU) to replicate single site findings
      • Exploratory analysis of cross site patient data (will add UCI data to cohort)
        • Waiting for control comparisons on cranial vault and brain volume, then we will look at the hippocampal volume in patients while checking out the other regions in controls
      • Preliminary assessment of possibility and utility of Polina’s shape analysis methods for this data set so we can define additional hypotheses to test
        • Not attempted yet.

2005.05.11 (C. Fennema-Notestine)

    • Abstract View – SFN submission 05/05:

Feasibility of multi-site clinical structural neuroimaging studies of legacy data: Aging & Alzheimer's disease C. Fennema-Notestine1*; R. Gollub2; B. Fischl2,3; B.T. Quinn2; J. Pacheco2; B. Dickerson2,4 1. Dept of Psychiatry, UCSD & VASDHS, La Jolla, CA, USA 2. Martinos Ctr for Biomedical Imaging, MGH, Charlestown, MA, USA 3. Computer Science & AI Lab, MIT, Cambridge, MA, USA 4. Dept of Neurology, MGH/BWH, Boston, MA, USA The Biomedical Informatics Research Network (BIRN) aims to enable scientists to conduct clinical imaging studies across multiple sites to test new hypotheses on larger cohorts. Given that many research groups have valuable existing (legacy) data, one goal of the Morphometry BIRN Testbed has been to assess the feasibility of pooled analysis of legacy structural imaging data. The present study aims to determine whether such legacy data can be meaningfully reanalyzed as a larger combined data set by using rigorous data curation and image analysis methods; in this case, to test the hypothesis that hippocampal volume decreases with age. Legacy T1-weighted MR and demographic data related to normal aging and Alzheimer's disease have been shared through the BIRN by UCSD (TL Jernigan; L Thal; D Salmon), MGH/BWH (M Albert; D Blacker; R Killiany), and Washington Univ. (R Buckner; J Morris). This preliminary report describes our work with older normal control data: UCSD (n=53/28F), MGH/BWH (n=36/22F), and WashU (n=49/24F). Cohorts from the sites were similar in age, education, and mental status. All MR data were analyzed in an automated manner with atlas-based Freesurfer segmentation software to generate volumetric measures of regions of interest (Fischl et al., 2002). Despite closely matched cohorts, mean hippocampal volumes differed across sites. Controlling for age and sex eliminated this site difference. Across the entire multi-site sample, age-related hippocampal volume loss was highly significant ( =-.42, t=-5.4, p<.001). This result suggests that legacy MR data from multiple sites can be pooled to investigate questions of scientific interest. Support Contributed By: Morphometry BIRN (http://www.nbirn.net), NCRR U24-RR021382; NIA P50 AG05131; DVA Med Res Svc; NIA PO1 AG 04953; ADRC/Howard Hughes; R01 RR16594-01A1; The MIND Inst; P41-RR14075

    • Next meeting scheduled for Tuesday May 17th.

2005.04.17 (C. Fennema-Notestine)

    • MGH testing adapted algorithm used to estimate intracranial volume (BD, BF).
    • QA for UCSD, BWH/MGH, and WashU underway.
    • Preliminary statistical review of normal control site effects underway (see Xythos site updates).

2005.04.11 (C. Fennema-Notestine)

    • Enhancement of the subcortical segmentation software (MGH) required manual segmentation of representative data sets for the creation of a training atlas, improvements of the skull-stripping and image registration algorithms, as well as adaptation of an algorithm to estimate total intracranial volume from the T1 images for normalization purposes (WashU/MGH).
    • We have just completed the subcortical segmentation of over 280 total data sets from the contributing sites. Our preliminary analysis shows that the previously significant site difference (UCSD vs. MGH/BWH) in hippocampal volume in the healthy control subjects was a consequence of the variations in image processing which we have now eliminated.
    • Final data curation of the clinical data (age, sex, diagnosis, education, and MMSE score) and subcortical analysis results are now underway.
    • Newly implemented 3DSlicer QA review tool (see 2005.03.10 notes below) is currently being tested at MGH and UCSD. The tool provides custom, efficient mode for reviewing large numbers of image sets, including the subcortical segmentation output.

2005.03.11 (C. Fennema-Notestine) MAD Update from MGH meetings March 9-11, 2005

    • Group co-leaders: Christine Fennema Notestine (CFN), Randy Gollub (RG), Brad Dickerson (BCD)
    • Based on work with: Bruce Fischl (BF), Brian Quinn, Tosa, Heidi, Nicole Aucoin, Dave Kennedy, Jorge Jovicich
    • Standard meeting plan CFN, RG, BCD: Tues 5:30EST/2:30PST
    • Completed Goals:
      • Finalized details of the participating sites and available data.
      • Defined details of the image processing steps, including QA proposals, for subcortical segmentation path.
      • Created tracking system by site for each data set.
      • Developed prototype segmentation review process within 3DSlicer
      • Determined action plan for project studies with abstract deadline goals.
    • Immediate Goals:
      • Review of current automated volumetric results by BCD and CFN for preliminary summary of outliers, site differences, etc.
      • Finalize exclusion criteria for initial phase (BCD, CFN)
      • RG and BCD to continue dialogue with BF to get estimated cranial vault measure from existing T1 data to control for individual differences in head size (eTIV: orig to MNI reg vs. aseg to template reg)
    • Future Goals:
      • Determine the status of all data sets that are to be included in the final analysis, minus those that don't meet QA standards) using 3DSlicer review process.
      • Define appropriate statistical analysis methods for the multi-site data.
    • Study Phases and Conference presentation goals:
      • Begin with cross-site study of control participants only on hippocampal volume with cranial vault estimates for proportionalized measures
        • To examine site effects
        • To replicate expected findings (Provide reference links)
          • Normal aging related volume loss
          • Gender effects and interactions
          • Potential education effects
          • Asymmetry findings
        • Conference goal SFN Nov. 2005 (Wash DC; abstract ddl MAY 12th, 2005)
      • Begin patient work based on control findings with only hippocampal volume with cranial vault estimates for proportionalized measures
        • Control for site effects
        • Replicate expected findings (Provide reference links)
          • Group comparisons revealing greater volume loss in converters and in AD
          • Relationship of volume loss to dementia severity (MMSE) within impaired groups
          • Relationship of volume loss to memory measures
            • MGH/BWH and UCSD CVLT
            • All sites with standardized scores (CFN/DS and REFS)
        • Conference goal AAN April 2006 (San Diego, abstract ddl TBD, usually near NOV 1st, 2005)
        • Other possible locations for patient work submissions:
          • ANA Sept. 2005 (San Diego, abstract ddl April 4th)
          • ADRD July 2006 (Madrid; abstract ddl TBD)
      • Combined multi-site study, all groups
        • HBM June 2006 (Florence, Italy; abstract ddl DEC 15th, 2005) for all subject combined study

2005.03.10 (N. Aucoin)

    • QA notes from the QA discussion held at MGH yesterday, with C. Fennema Notestine, B. Dickerson, N. Aucoin

2005.03.01 (C. Fennema Notestine)

    • UCSD data has been uploaded into BIRN SRB, and downloaded locally for CFN to review. First pass review of data underway.
    • UCI has updated the dataset descriptions available on the MAD website.
    • March 8-11 working group at MGH (Randy, Brad, Bruce, Christine, etc.). More details of specific work plan following MBIRN AHM meeting in Miami (March 2-4).

2005.02.23 (R. Gollub)

    • See MBIRN Xythos site (https://share.spl.harvard.edu/xythoswfs/webui ) for posting of the updated excel file that now includes the neuropsych page thanks to Christine.
    • Each of the sites (MGH/BWH, UCSD, Wash U, UCI) that are interested in participating need to collect and enter the requested information in advance of our meeting in Miami. Initially a y= yes or n= no response is acceptable so that we know what data is available for sharing, but by the time of the meeting it is critical to have all fields completed. To date BWH/MGH and UCSD have completed the y/n part and are working on filling in the fields.
    • Randy G has been in contact with Steve P and Jess T of UCI. The dependencies for their participation have been discussed and are currently being resolved.
    • Randy G has just contacted Randy B of Wash U to get them on board.
    • Randy G, Steve P, Silvester C., and Shawn M are meeting tomorrow 2/24 to discuss technical developments of 1) a GUI front end to 3D Slicer and 2) generation of movies to facilitate rapid and precise review of the image data as it flows through the Freesurfer processing stream.


2005.02.17 (C. Fennema Notestine) Action Items for General Goals:

        • Dataset updates:
          • UCSD data sharing confirmed, with scientific and authorship involvement at general level to include CFN, Terry Jernigan & Leon Thal on all projects, David Salmon and Greg Brown on a by-project basis; CFN will review data after HS @MGH uploads to SRB.
          • BWH/MGH data sharing being confirmed, with scientific and authorship involvement at general level to include BD, Deborah Blacker, Marilyn Albert, & Ron Killiany on all projects; BD is reviewing MGH data.
          • UCI data sharing being discussed with Steve Potkin (RG)
          • WashU data sharing being discussed with Randy Buckner (BD)
        • Technical work with analysis team:
          • finalize conceptual QA priorities & specific pipeline
          • define inclusion/exclusion criteria (i.e. based on clinical measures vs. scan measures)
        • Define initial scientific clinical aims and sites included in first steps (will create aims doc for sharing site)
          • Design of statistical analysis path for scientific clinical aims
          • Plan goals and dates for CFN visit to MGH
            • Possibly technical aims related to QA
            • Possibly refine specific aims and plan statistical design
            • Possibly statistical analyses across relevant sites
        • Action Items for Specific Sites:
          • MGH data (BD): continue discussion with Deborah re: sharing of BWH/MGH data. Confirm number of available scans that can be pooled and compare with those processed by Jenni. If different, provide Jenni information for additional processing. Arrange QA of the BWH/MGH data that Jenni already processed.
          • WashU data (BD): Coordinate with Brian Q. the WashU datasets (numbers, processing software version, QA, clinical data) and what data is to be shared for this project
          • UCI data (RG): set up call with UCI (Dave Keator, Jess Turner, Steven Potkin) to clarify permissions and use of this data for publication. Clarify technical issues re preprocessing (defacing, formats, image orientations) prior to data arrival to MGH. Establish points of contact for sharing clinical and neuropsych metadata.
          • UCSD data (CFN): review data.

2005.02.11 (R. Gollub) Project update 1) We have a project folder on the MBIRN Xythos site (https://share.spl.harvard.edu/xythoswfs/webui ). In it is an Excel file (see all 4 pages) that we will use to capture all the relevant information to proceed smoothly. It currently contains a fairly complete entry for the BWH/MGH data set (thanks to Brad).

      • IMPORTANT NOTE: the entry of information REQUIRES that you have obtained permission to share the data with BIRN for this multisite study. There is no problem if you must publish a single site result in parallel (See below). For the first pass, it is sufficient to indicate for certain fields a binary Yes or No for availability, knowing that we will soon be seeking to replace this with actual information.
      • Each time you are ready to enter new info to the EXCEL file, check the xythos site to get the most recent version and as soon as you are done, upload your revised version with a new date.

2) We have determined that multiple of the sites expect to publish the results of their own data from this analysis in addition to the shared multi-site analysis (e.g. BWH/MGH and UCI, others?). That is great. More papers more better. Clarification of these plans will greatly facilitate our planning and prioritization of the work. 3) We have laid out the plan for "locking" the version of Freesurfer (and all it's component binaries and libraries) that is required for the analysis so that it will be the same for all sites. We are still confirming that the current version is going to work for the UCI data and for the most severe of the AD cases from each site. 4) Action Plans:

      • a) Christine- we need you to download the Excel file and add all the neuropsych data types from the initial AD Demo as well as to fill in the fields for the UCSD data.
      • b) Jorge- please get the DICOM header files from Brian to fill in the scan parameters for the Wash U and UCSD data and enter in the information.
      • c) Once done, we will pass the Excel file to Randy Buckner to determine who his study contact will be and get his site information completed.
      • d) David/Jess- we will pester you (or your designee) to fill in UCI stuff once Christine is done, but you need to be preparing to enter all the UCI information. It is not possible for anyone outside the study acquisition staff to know exactly what was done to collect the clinical and neuropsych data or exactly how the answers were coded, so we really need a contact person.
      • e) Any one who notices information I have forgotten, please add it!

5) By the Miami meeting we need to have this EXCEL file complete so that our shared time together can be used to devise an implementation plan with timeline. Topics I suggest we have finalized by the end of the Miami meeting:

      • a- criteria for data sharing agreed upon
      • b- plans for single site publications clarified
      • c- specify clinical and neuropsych fields available in common across sites
      • d- specify the point person and work needed per site to pass the data into this pooled analysis
      • e- work with the data integration team to get resources for data management for this project (read our own instance of XNAT/Oasis)
      • f- MOST IMPORTANTLY- From this we should be able to estimate what our final pooled cohort will look like (e.g. approximate n's and patient groups) and complete the formulation of our multi-site study hypotheses.

6) Christine is planning a trip to Boston shortly after the Miami meeting to turbo charge our forward progress. Suggested goals for Christine's visit can be refined in Miami, but currently include:

      • a- specify QA criteria and plan for image quality
      • b- help finalize the tools for the QA step (Slicer to view MPEGs of the asegs overlaid on norm data)
      • c- specify minimal clinical and neuropsych data
      • d- meet with Dave/Heidi to advance specifications for Query tool development
      • e- start the actual analysis!


2005.02.03 (J. Jovicich)

      • Brad Dickerson, Bruce Fischl, Randy Gollub, Dave Kennedy & Jorge Jovicich met on Jan 31 to review status and plans for pooling multi-site subcortical and clinical information of retrospective data
      • Available datasets are (needs confirmation):
        • BWH/MGH: ~50 old controls, ~100 MCI; ~30 AD
        • UCI: ~500 MCI
        • UCSD: ~55 old controls; ~12 MCI; ~60 AD
        • Wash U: ~175 old controls; ~175 AD
      • Pending processing:
        • QA UCSD asegs, upload to SRB for UCSD
        • QA BWH/MGH asegs
        • WashU analysis status update (asegs with latest version, QA)
        • UCI QA before arrival to MGH, UCI analysis, QA, upload to SRB for UCI
      • Action Items
        • Brad: will set a call with Christine to follow up & plan her visit as well as how to best prepare for her visit. Discuss potential paper goals and authorship issues. Double check with Deborah that we can use the BWH/MGH MRI/clinical data. Ask for conversion rates. Confirm number of available scans that can be pooled and compare with those processed by Jenni. If different, provide Jenni information for additional processing. Arrange QA of the BWH/MGH data that Jenni already processed. Coordinate with Brian Q. the WashU datasets (numbers, processing, QA, clinical data)
        • Randy: set up call with UCI (Dave Keator, Jess Turner, Steven Potkin) to clarify permissions and use of this data for publication. Clarify technical issues re preprocessing (defacing, formats, image orientations) prior to data arrival to MGH. Establish points of contact for sharing clinical and neuropsych metadata. This meeting will happen on Fri Feb 4. DONE (R. Gollub)

For more information see: