@granvillejmath
Enthusiast, dad, researcher. Working at Karolinska Institutet with PET quantification and analysis. Interested in (Bayesian) stats, molecular imaging, pharmacokinetics, #rstats, brains and brain imaging. Scottish South African in Sweden.
If you have any questions or would like some help with implementing the model, don't hesitate to shoot me a mail! We also have a clinical paper coming soon with this data to show that this model can identify new clinically-relevant associations which were not previously identifiable!
This model is currently implemented in #rstats, @mcmc_stan and #brms, but it should be quite straightforward to translate to other languages. I've included the code and sample data, as well as a fully documented notebook to demonstrate how to use it github.com/mathesong/Si...
It is known that there is regional variation in the rate of decreases in serotonin 1B receptor BP. So we checked the variations from the mean age effect, and even here, our outcomes were highly replicable.
Despite all these differences, our model yielded highly similar estimates of the association between model parameters and age across datasets. We also found that we could also successfully combine and harmonise datasets in the same model using covariates to improve performance further.
Applied to real datasets, we decided to test how replicable our mode's inferences were in data from three different PET centres, with different PET systems, acquisition protocols, preprocessing, subject demographics. So we looked at the effects of age on regional serotonin 1B receptor availability.
We also show large improvements to inferences made. Here, estimating simulated treatment effects relative to a placebo intervention. A: large increases in power, and B: our model's estimates of treatment effects are closer to the true value across simulated datasets
We show dramatic improvements in quantification using SRTM, reducing the RMSE by over 50% on average (and up to 90% for one region) for BPND, but also for the other estimated parameters too.
This builds on our previous work where we introduced this approach, called SiMBA, for gold-standard invasive PET quantification i.e. with arterial blood collection. But most quantitative PET uses reference regions instead: now SiMBA can be applied for this too!
www.sciencedirect.com/science/arti...
Really excited to announce our latest preprint. We introduce a new approach for joint Bayesian quantification and analysis of PET data using reference tissue models. It improves quantification, inferential efficiency (power) and also allows efficient harmonisation across PET systems or centres.
Ah, the to-do list pile got a bit too high and I've dropped off the #adventofcode train... Will see if I have time to join later in the month, but Godspeed to the rest of you hanging on, especially team #rstats !
Day 5 of #AdventOfCode with #rstats : "Print Queue" (adventofcode.com/2024/day/5). Not pretty, not elegant and not particularly quick -- but it works! This one was quite fun: I can't remember the last time I used a while loop!
Riiight - nice idea! All that lagging and leading is going to get pretty brutal though! 😄
Mixing up a bit of base and a bit of tidyverse whenever they feel right just suits the way my brain works - but all the really "paradigm"-consistent solutions always just look so satisfying!
Day 4 of #AdventOfCode - "Ceres Search" with #rstats. I found this one pretty tricky! Had to dig into the old base R toolbox. Really interested to see how this could be done in a more #tidyverse way.
As a regular MCMC user, this hits close to home. Painful memories of calculating our simulation run-times for the cluster in core-years!
After seeing some of you folks rocking the cheeky eval(parse()) today, I'm feeling extra inspired to follow along and learn some new tricks... I'll try to keep up for as long as I can!
And thanks so much to @riinu.bsky.social for teaching me about carbon.now.sh for generating the geautiful code figures!
Day 3 of Advent of Code 2024: "Mull It Over" with #rstats . As bad of a reputation as it gets, I do love me some regex!!
#AdventOfCode
gist.github.com/mathesong/35...
Awesome - thanks!! I've seen these so often and always wondered...
I love your clean code screenshots! Are you just taking screenshots of your editor, or is there some package or add-on you're using for generating these?
My #rstats solution to "Red-Nosed Reports" - Day 2 - Advent of Code 2024 #AdventOfCode adventofcode.com/2024/day/2
gist.github.com/mathesong/78...
In the spirit of making a new home here, and re-kindling the fun and community of Twitter of yesteryear, I'll join in for #AdventOfCode this year (adventofcode.com/2024/day/1) (at least as many as I can find time for).
Here's my #rstats solution for Day 1: gist.github.com/mathesong/ec...