Thread by @rai_bethlehem

“Brain charts for the human lifespan”! The result of 200 people putting their 🧠🧠🧠together (120,000+ of them!) to generate growth curves from mid-gestation to 100 years. brainchart.io & biorxiv.org/content/10.110… @jakob_seidlitz @SimonWhite83 @edbullmore @Aaron_A_B

How did we get here? Through countless zoom hours, DUAs and emails, we aggregated data from over 100,000 individuals across 6 continents, comprising over 120,000 structural brain scans and racked up 1,400,000 @FreeSurferMRI computing hours processing them @CambridgeRcs.

Next we planned to run these derived variables through @WHO recommended GAMLSS models and quickly found we had to go a little beyond the standard functionalities if we were to account for the complexity posed by multi-site imaging data.

Luckily @SimonWhite83 from the @MRC_BSU and @psychiatry_ucam was up for a challenge and helped generate customised GAMLSS to map the non-linear trajectories of brain volume changes across the lifespan, and their variance and rate of change too!

Since our collective dataset also contained imaging from over 20 psychiatric or neurological cohorts, many with hundreds of scans, we could investigate whether these “growth charts” could provide some clinical insight or stratification.

These charts allow standardised assessment of someone's “centile” score, relative for their age & sex. This captured known patterns of tissue loss in neurodegeneration and revealed surprisingly large effect sizes in some psychiatric conditions.

Of course this was all from cross-sectional data and so one has to wonder how stable these centiles really are? Luckily we also had around 20,000 longitudinal scans to test that, since you can’t scan someone across their lifetime :) . Yet.

As it turns out, the centiles are pretty stable! In control subjects these scores only fluctuate on average only about 5% and in some clinical cohorts even less.

Now you might be thinking, “what about my data? My data wasn’t in this model, but I also want some of these cool normalised centile scores…”

There are too many people to thank or even track down on twitter but this would not have been possible without a massive collaborative effort and emails, slack messages and zooms between @rai_bethlehem @jakob_seidlitz @Aaron_A_B @_JakeVogel_ @SimonWhite83, @KevinMAnder

@edbullmore @psychiatry_ucam @ARC_Cambridge @CHOP_Research @Penn_SIVE @PennPsych @Cambridge_Uni and countless contributors many of whom not on twitter @sattertt @takishinohara
@sbaroncohen @DrDamienFair @misicbata @bttyeo
@AndrewZalesky @sylv_villeneuve @RikOssenkoppele
...

@hlschaare @sofievalk @Warrier_Varun @hollabharath @vdcalhoun @mvlombardo @timrittman @zuoxinian @SimonWarfield @katjaQheuer @R3RT0 @alexa_pichetb @_michael_scholl @OskarHansson9 @heathersibley99 @HZiauddeen @gadevenyi

@rikhens @DuncanAstle @sophieadler
@KonradWagstyl @bogglerapture @petra_vertes
@mallarchkrvrty1 @PaulPcf22 @saashi_bedford @rafa_romero_gar @VandyAtVandy @gsalumjr
@nialljbourke @JamesCole_Neuro @maggieeastfire
@wakeworksleep @garedaba @Sarah_Morgan_UK

As well as data from (amongst many others): @camcan_tweets @calmcbu @HumanConnectome @DevelopingHCP @meld_project @MGHMartinos @BostonChildrens @uk_biobank @NIMHgov any many many many others not on twitter.

Supported by: @BritishAcademy_ @The_MRC @NIH @theACEcharity @CambridgeBRC countless funders contributing to one or more of the many datasets included in this project and of course all the volunteers contributing their time and effort to be scanned as part of a research project!

And I think brainchart.io and github.com/ucam-departmen… would entitle us to throw a #OpenScience in here as well

Turns out you can! We developed an out of sample estimator to generate these estimates on novel datasets (recommended minimum N~100) and made it interactively available through #RStudio #shinyapps brainchart.io