Towards large-scale brain imaging studies: How to deal with analytic variability? April 19th, 2018 Camille Maumet
Outline Introduction: VisAGeS and AI Large-scale brain imaging studies Analytic variability 2 Camille Maumet - AI in our labs April 19th, 2018
Introduction VisAGeS and AI
VisAGeS research objectives Understand the brain under typical and pathological conditions with brain imaging Team leader : Christian Barillot Goals ● Early Diagnosis ● Therapeutic choices ● New therapeutic protocols Multiple sclerosis, Psychiatry, Parkinsonian disorders, Dementia, Stroke (Slide content from Christian Barillot, adapted) Camille Maumet - AI in our labs April 19th, 2018
Multiscale «Brain Imaging Biomarkers» 5 Majors challenges • From Bench to the Bed • Models and algorithms to reconstruct, analyze and • Mass of data to store, distribute and “semantize” transform • From ms to Century ( 3*10 12 ratio) Contributions & skills • Model Inference • Data fusion ( multimodal integration, registration, patch analysis, … ) • From nm to m ( 10 9 ratio) • Statistical Analysis & Modeling • High dimensional optimization • Sparse Representation ( compressed sensing, dictionary • Data integration learning ) • Brain computer interface • Machine Learning ( supervised/ unsupervised classification, • … (Slide from Christian Barillot, adapted) discrete model learning ) Camille Maumet - AI in our labs April 19th, 2018
VisAGeS in AI Applied neuroimaging AI methods Neuroimaging methods Camille Maumet - AI in our labs April 19th, 2018
Towards large-scale brain imaging studies
Sample sizes in brain imaging research 2015: 30 subjects / study [Poldrack et. al, Nature Neuroscience 2017] 8 Camille Maumet - AI in our labs April 19th, 2018
Sample sizes in brain imaging research 2015: 30 subjects / study Low diversity & Low statistical power [Poldrack et. al, Nature Neuroscience 2017] 9 Camille Maumet - AI in our labs April 19th, 2018
More and more open data are available! Photo de Neil Conway Single study 30 subjects Consortium 1000 subjects + Images Cohort + Homogeneous 1 000 - 100 000 subjets - Fewer 10 Camille Maumet - AI in our labs April 19th, 2018
How to deal with analytic variability?
Challenge: analytical variability Raw data Feature extraction Derived data Statistical analysis Results 12 Camille Maumet - AI in our labs April 19th, 2018
Challenge: analytical variability Raw data Raw data Feature extraction Derived data Statistical analysis Results Camille Maumet - AI in our labs April 19th, 2018
Challenge: analytical variability Raw data Raw data Feature extraction Feature extraction Derived data Derived data Statistical analysis Results 14 Camille Maumet - AI in our labs April 19th, 2018
Challenge: analytical variability Raw data Raw data Feature extraction Feature extraction Feature extraction Derived data Derived data Derived data Statistical analysis Results 15 Camille Maumet - AI in our labs April 19th, 2018
Challenge: analytical variability Raw data Raw data Feature extraction Feature extraction Feature extraction Derived data Derived data Derived data Statistical analysis Statistical analysis Results Results Meta-analyses 16 Camille Maumet - AI in our labs April 19th, 2018
Quantify Compensate Estimate variations Remove unwanted across pipelines "pipeline effect" 17
Quantify Compensate Estimate variations Remove unwanted across pipelines "pipeline effect" 18
Impact of Analysis Software on Task fMRI Results ● 3 published studies ● Reanalysed with 3 fMRI tools ● Reusing the same data Research question: how choice of software package impacts on analysis results? 19 Camille Maumet - AI in our labs April 19th, 2018
Impact of Analysis Software on Task fMRI Results Reproducing the main figure Study 1 Study 2 Study 3 Preprint : Bowring, Maumet* and Nichols*, 2018. www.hal.inserm.fr/inserm-01760535 April 19th, 2018 20 Camille Maumet - AI in our labs
Impact of Analysis Software on Task fMRI Results Reproducing the main figure Dice coefficients: 0.23 - 0.38 Study 1 Study 2 Study 3 Preprint : Bowring, Maumet* and Nichols*, 2018. www.hal.inserm.fr/inserm-01760535 April 19th, 2018 21 Camille Maumet - AI in our labs
Impact of Analysis Software on Task fMRI Results Reproducing the main figure Dice coefficients: 0.23 - 0.38 Study 1 Study 2 Study 3 Unthresholded statistics Preprint : Bowring, Maumet* and Nichols*, 2018. www.hal.inserm.fr/inserm-01760535 April 19th, 2018 22 Camille Maumet - AI in our labs
Impact of Analysis Software on Task fMRI Results ● Challenges ○ Use the "same" pipeline across fMRI tools ■ Implementation details ↔ Methodological differences ○ How much difference is too much? ■ "Compatibility" across pipelines 23 Camille Maumet - AI in our labs April 19th, 2018
Quantify Compensate Estimate variations Remove unwanted across pipelines "pipeline effect" 24
2. Remove unwanted "pipeline effect" Raw data Raw data Feature Feature extraction extraction Derived data Derived data Recalibration Statistical analysis Results 25
Camille Maumet Photo de Neil Conway Towards large-scale brain imaging studies: How to deal with analytic variability? Camille Maumet - AI in our labs April 19th, 2018
Recommend
More recommend
