NIH MRI Study of Normal Brain Development AC

NIH MRI Study of Normal Brain Development AC Evans Ph. D. Brain Development Cooperative Group Pediatric Functional Neuroimaging: a Trans-NIH Workshop May 25, 2004

Contrast changes over time 2

Problems with previous studies • • Sample sizes too small to detect subtle signals Heterogeneity of subject population Little longitudinal data Lack of demographic representativeness Limited behavioral data for brain-behaviour correlation Limited MRI data (typically T 1 only) Usually limited analysis techniques 3

MRI Study of Normal Brain Development (N=500) Create a database of behavioral and brain MRI development data for 0 -18 years Analyze structural-behavioural relationships Develop technique for dissemination of results The National Institute on Drug Abuse 4

Rationale for Project Design • Problem: Existing normative databases limited in size Pooling of databases difficult. Existing databases incompatible in – Slice thickness – Pulse sequence – Demographics – Behavioural tests One centre cannot collect large dataset fast enough to keep pace with technology • Solution: Clinical trial model: multi-centre acquisition, uniform protocol 5

MRI Objectives • Objective 1: Anatomical MRI/Behavior (5 -18) • Objective 2: Anatomical MRI/Behaviour (0 -4) • Ancillary A: MR Spectroscopy • Ancillary B: Diffusion Tensor Imaging, Relaxometry 6

Pediatric Study Centers (PSCs) • Neuropsychiatric Institute and Hospital, UCLA Mc. Cracken • Children’s Hospital, Boston Rivkin • Children’s Hospital of Philadelphia Wang • University of Texas-Houston Medical School Brandt • Children’s Hospital Medical Center, Cincinnati Ball • Washington University, St. Louis Mc. Kinstry 7

Data Coordinating Center (DCC) • • • Overall Direction Database Behavioral Liaison MRI Acquisition MRI Analysis Sampling Plan Data Transfer Scientific Liaison Clinical Liaison Evans Zijdenbos, Vins, Charlet, Harlap, Das Leonard, Milovan Pike, Arnaoutelis Collins, Kitching, Lerch Lange (Harvard) Zeffiro, Van Meter (Georgetown) Paus Ad-Dab’bagh, Webster 8

Clinical Coordinating Center (CCC) – St. Louis Recruitment, behavioral measures selection/certification, exclusions etc. for Obj 1, 2 – Botteron, Almli Behavioral QC – Rainey, Henderson, Singer, Smith, Dubois, Warren, Edwards DTI Processing Center (DPC) - NIH Pierpaoli, Basser, Rohde, Chang MRS Processing Center (MPC) – UCLA (? ) Alger, O’Neill 9

NIH MRI Study of Normal Brain Development DCC CCC DPC 10

Recruitment Procedure • • • Representative sample based on US 2002 census Zip code demographic data Telephone brief screener at recruitment Telephone long screener for inclusion criteria DISC, FIGS, CBCL Hospital Visit (Neuro exam, Behaviour, MRI) Objective 1 scans 3 times, every 2 years Objective 2 scans 3 -6 times SES (3 levels) X age (0 -18 yrs) X gender X ethnicity 11

Accrual by Age (Objective 1) 60 50 40 30 20 10 0 4. 5 6 7 8 9 10 11 12 13 14 15 16 17 18 12

Family Income (Raw) N = 409 13

Parental Education N = 409 NIH MRI Study of Normal Pediatric Development US Population 14

Ethnicity N = 409 15

Behavioral Maturation is multi-dimensional 16

Test Battery MRI BEHAVIORAL T 1 W (Obj 1+2+fallbacks) Full Interview T 2 W (Obj 1+2+fallbacks) Bayley Mental Handedness PDW (Obj 1+2+fallbacks) Bayley Motor JTCI MRS (Obj 1+2) Bayley Behavioral Nepsy fluency MRSI (Obj 1+2) Brief Interview Neurologicals DTI (Obj 1+2) BRIEF Parent PSI Dual-contrast T 2 (Obj 2) CANTAB Pregnancy T 1 Relaxometry (Obj 2) Carey Purdue peg board T 2 Relaxometry (Obj 2) CBCL Tanner CVLT (C and II) TCI DAS Urine and Saliva DISC Digit span and coding DPS 4 WASI Exclusion (Obj 2) FIGS Woodcock-Johnson III 17

Study Organization Study Subject n-1 Subject n Visit 1 Exclusionary CBCL Screening DISC Behavioral JTCI Instruments CANTAB MRI Procedures MRI Visit 2 Subject n+1 … Visit n CBCL DPS 4 WJ 3 WASI MRI 18

System Architecture PSC DCC BVL MRI Console Behavioral PC (laptop) RNE MRI Study Work Station INTE MRI Scanner T Mass Storage System BVL Internet & DBMS Server(s) MRI Data Warehouse Backup System Data Marts BVL Scientific Community Data Analysis Pipeline 19

New technology never works first time 20

DCC-ID Candidate Profile identified by PSCID for each candidate personal contains data on multiple visits Session. ID recruited by Gender Visit. No ethnic psc Do. B visit Objective member of Ethnic. ID Weight Age stores data for a battery of administered MRI procedures & behavioral instruments MRI procedures Center. ID Height Objective behavioral battery of instruments DICOM T 2 W 3 D bio figs apib das tanner cantab psi MRS exclus disc carey neuro wasi cvltc purdue waisr header MRSI brief int dps 4 hand pls 3 wj 3 cvlt 2 saliva T 1 W 3 D PD full int cbcl nepsy pregn bayley jtci urine Type Screening wisc MINC Objective. ID are identified by Test ID Test. ID Comment. ID Score. ID 21

DBMS Software Platform • My. SQL DBMS: - Cross-platform, open source - Robustness, speed, reliability - Low development cost - Remote management • Graphical User Interface: - Cross platform, Internet enabled application - PHP-based application, complemented by Java. Script, Java, and Perl for data manipulation tasks. - Remote management, customizable 22

Database GUI 23

24

Database Summary • • • Low-cost, extensible, secure 61 tests, approx. 20, 000 possible data fields (~1000 filled/subject) Laptop-based behavioral test battery Automatic MRI data transfer Web-based behavioral GUI Interactive 3 D MRI web viewer Automatic QC procedures Project web site 25

N=449 N=56 N=188 Obj 1 Obj 2 DTI 26

Behavioral Instrument Status Total # of Instruments = 9827 Objective 1 = 9029 Objective 2 = 798 Total # in Database = 9827 As of May 20, 2004. 27

IQ Scores (n=248) 40 30 20 10 Std. Dev = 12. 50 Mean = 110. 6 N = 248. 00 0 75. 0 80. 0 95. 0 90. 0 105. 0 100. 0 115. 0 110. 0 125. 0 120. 0 135. 0 130. 0 145. 0 140. 0 FIQ: 110. 7+0. 8 VIQ: 109. 9+08 PIQ: 108. 9+0. 8 28

Summary of Behavioral QC SITE (N) ERRORS [Error rate = (Errors/1, 000 x. N) x 100] INPUT (%) SCORING (%) CLERICAL (%) QUERIES (%) OTHER (%) BOS 1 (29) 133 (0. 4) 3 (0. 01) 213 (0. 7) 13 (0. 05) 19 (0. 06) BOS 2 (4) 3 (0. 07) 0 (0. 0) 14 (0. 4) 5 (0. 1) 4 (0. 1) CIN (24) 116 (0. 5) 3 (0. 01) 108 (0. 4) 1 (0. 004) 15 (0. 06) HOU (23) 126 (0. 5) 7 (0. 03) 233 (1. 0) 5 (0. 02) 15 (0. 06) UCLA (18) 46 (0. 2) 1 (0. 005) 155 (0. 9) 4 (0. 02) 8 (0. 04) PHI 1 (14) 71 (0. 5) 1 (0. 007) 51 (0. 4) 0 (0. 0) 4 (0. 03) PHI 2 (1) 0 (0. 0) 2 (0. 2) STL 1 (29) 140 (0. 5) 4 (0. 01) 106 (0. 4) 12 (0. 04) 41 (0. 1) STL 2 (12) 99 (0. 8) 0 (0. 0) 86 (0. 7) 21 (0. 2) 17 (0. 1) Total (154) 734 (0. 5) 19 (0. 01) 966 (0. 6) 61 (0. 04) 125 (0. 08) 0 (0. 0) 29

W-J: Passage Comprehension (n=278) r=0. 85, p=0. 000 30

WASI: Vocabulary (n=248) 31 r=0. 86, p=0. 000

Spatial Working Memory (CANTAB): Errors (n=250) 32 r=-0. 75, p=0. 000

Objective 1 33

Objective 1 – MRS/I Objective 1 – DTI 34 *Corrected to exclude the early Cincinnati and St. Louis 1 subjects since the DTI product was not available.

Objective 2 35

QC Overview Number of: Oct 2003 May 9, 2004 MRI datasets received: 492 (obj 1 + obj 2) 582 (436 obj 1 + 146 obj 2) MRI datasets QC’d 484 (98. 4%) 582 (100%) Total volumes rec’d 8815 13102 Time spent in MRI QC/subject 2 -100 minutes unchanged Total time spent in MRI QC ~360 hours ~530 hours (total) Ø Goal: quick turn-around time – mean time for expedited review – median time over all subjects 1. 5 days 7. 0 days Ø Concentrated mostly on subject QC 36

Inter-packet movement After Before • Separate volume into packets • Register each packet to target • Resample and interpolate to 1 mm slice thickness 37

Data Flow for Brain Mapping Data Acquisition : Reconstruction : Conversion to MINC Image Format PET f. MRI Registration a. MRI - e. PET a. MRI - t. PET Frame alignment Registration a. MRI - f. MRI Partial volume correction Voxel-based model fitting Voxel-based coherence analysis a. MRI Inter-slice normalization T 1/T 2/PD/… alignment Intensity non-uniformity correction 3 D segmentation Stereotaxic Spatial Normalization Functional Probability Maps Inter-volume normalization GLM analysis in 3, 4, or 5 D Structural Probability Maps 38

Anatomical MRI analysis pipeline ASP 39

INSECT ANIMAL ASP SEAL Manual 40 Auto

Objective 1 – classification 41

Obj 1 – Tissue SPAMs (n=337) 42

Age-related changes in WM density 43 Paus et al, Science 1999; n=111 NIHPD; n=204; 16 1 0 t=10. 5

WM density and Spatial Working Memory: Between Errors (Age removed) NIHPD; n=188; 7 -1 66 t=-4. 0 44

Cortical Surface Extraction (Kim, Hanyang U. ) 45

Cortical Surface Extraction (Kim, Hanyang U. ) 46

Automated extraction of both cortical surfaces using CLASP algorithm (5 different brains) 47

Analysis of detection limits • 19 T 1 MRIs of the same subject (Colin Holmes) with 1 mm isotropic sampling. • Computation of standard deviations across cortex. – Across blurring kernels. – Across metrics. • Power analysis: – N needed to recover change of 0. 5 mm – Change required at N=25 in each group. 48

Colin’s 19 Brains 49 Average

Required N to recover 0. 5 mm Unblurred 5 mm 10 mm 200 0 50

Recoverable change when N=25 Unblurred 5 mm 10 mm 2 mm 0 mm 51

Prefrontal atrophy in normal aging (N=851) 0. 025 Slope (mm loss) 0. 01 52

Cortical thickness vs. age between (Obj 1, 4 -18) (slope in mm/yr, N=289) 53

Cortical thickness versus Age 54

55

56

Obj 2 stereotaxic T 1 average nihpd 120_obj 2 icbm 152 57

Related large-scale projects • • • ICBM (7000) Giedd and Rapoport (3000) Brad Peterson (TS, OCD, ADHD 600) Maternal Adversity (MAVAN 500) Tourette’s Neuroimaging Consortium (500) Japanese Human Brain Project (1200) 58

Welcome to the good ship NIHPD 59