I prerecorded the audio below (8 minutes) to go with these PDF slides for AD/PD 2023 in Gothenburg, Sweden.
Disease Progression Modelling on MRI data Identifies Subtypes with Cognitive Heterogeneity in A4 Study Preclinical Trial Cohort
TL;DR: We used SuStaIn to find MRI-based subtypes in the screening data of the A4 Study clinical trial (preclinical Alzheimer’s disease) and leveraged the ADNI observational dataset to link these subtypes to heterogeneous cognitive decline.
This could potentially wash out any treatment effect (if present) in the A4 study trial. This work was completed before the results of A4 were known.
Posted on2021-04-30|Comments Off on New paper on data-driven progression subtypes in Alzheimer’s disease
Latest paper is an amazing collaboration led by the wonderful Jake Vogel that pulled together the largest database so far of tau PET imaging of Alzheimer’s pathology in the living brain, then unleashed Alex Young’s SuStaIn algorithm to discover four previously unknown/uncharacterised subtypes.
It turns out that these are quite common (as in not rare), and they have unique symptom profiles too.
I could dig in, but here’s what Jake had to say on twitter:
Does Alzheimer’s disease pathology spread through the brain consistently across the population? We delve into this in our paper “Four distinct trajectories of tau deposition identified in Alzheimer’s disease”, now online at Nature Medicine. https://t.co/eNAZ63pe3z Thread below 1/
Tau is one of the two pathologies that characterize Alzheimer’s disease (AD). Autopsy work suggests it spreads in a specific and highly compelling pattern, formalized into the Break staging system and used to stage tau pathology in AD. See https://t.co/hiKLt7hGY6. 2/ pic.twitter.com/y1LBXUIGPK
Parsing AD heterogeneity in an unbiased manner has been challenging because overall severity is the greatest source of spatial variance. Alex Young created the SuStaIn algorithm https://t.co/txD9ivcTKC, which handles both pseudotemporal and spatial aspects of variation 4/ pic.twitter.com/RB1WiojfPc
But this requires pretty large sample sizes. So, we got together with researchers at the @UCSFmac, @biofinder_study in Sweden, Gagnam Severance in Korea, AVID radiopharmaceuticals, + ADNI data to come up with over 1600 tau-PET scans. 1143 were used for analysis. 5/
Applying SuStaIn to this dataset, we found four distinct spatiotemporal patterns, including limbic-predominant and MTL-sparing variants similar to those described by @DrNeuroChic, as well as posterior and lateral temporal subtypes resembling the clinical variants, PCA & lvPPA 6/ pic.twitter.com/HhgflhbyjP
Each subtype was found in each of the five cohorts we studied. We then applied SuStaIn separately on another cohort, BioFINDER II, which used a totally different tau-PET radiotracer. Performing the analysis from scratch in this group, we found highly similar subtypes. 7/ pic.twitter.com/gMr4lGgShw
The 4 subtypes differed in their clinical presentation, including in terms of age, likelihood of APOE4 allele carriage, and cognitive profiles. We found the posterior subtype to exhibit a slower rate of global decline, while the lateral temporal variant progressed faster. 8/ pic.twitter.com/HNvsmXRYsF
Using network diffusion methods from our previous paper https://t.co/YLWgPETKd6, we found the tau-PET patterns of each of the 4 subtypes resembled a different temporal lobe network. Simulating diffusion through that network could recapitulate the subtype patterns observed. 9/ pic.twitter.com/tNdBtVslGE
In all, we conclude that there may not be such an entity as “typical AD”, but rather that individuals with AD present with one of at least four different subtypes with distinct prognoses. In addition, younger age is associated with more severe subtype expression. /10 pic.twitter.com/GLAqtmjrst
We are interested to see our findings validated by pathologists, and we would like to better characterize these subtypes. Will they respond differently to treatment? What does it mean about disease biology? Much work to be done. 12/
I have a vacancy for a post-Masters / pre-PhD level person interested in medical image (MRI) data wrangling and analysis, including computational modelling of neurological diseases like Alzheimer’s and Lewy Body Disease.
One important goal for the project is to use existing computational methods (from the POND group) to build a differential diagnosis tool for dementias, Specifically Alzheimer’s Disease vs Lewy body Dementia.
This is exciting for two reasons in particular: we are looking at real world data from individuals in the prodromal stage of dementia.
If you’re not quite ready for a PhD and are interested in medical image computing, this could be the role for you. The ideal candidate would have a background in computer science, physics, maths, medical image computing, or a related field.
Comments Off on Vacancy: Research Assistant in Memory Clinic Image Computing