MICCAI 2021#

../../_images/MICCAI_2021.png

Fig. 7 MICCAI 20201.#

Workshop abstract

The proposed tutorial is intended to present the most advanced and mature data-driven models for the modelling of neurological disease progression within a day-long session. The main objective is to illustrate the major challenges of modelling neurodegenerative disorders, especially the unknown (and heterogeneous) disease time axis, and, the reconstruction of long-term disease history from short-term individual observations - challenges existing beyond neurological applications. To that end, the tutorial will dive into state-of-the-art models that allows to take the best out of cross-sectional and longitudinal data. Short lectures introducing the intention of each DPM model will be followed by hands-on session based on Python notebooks that illustrate the different concepts in the context of Alzheimer’s disease progression. At the end, the participants will be able to select a DPM model suitable to their own dataset and implement it thanks to the software presented.

Conference workshop Hands-on session Full day

Link to the conference

Learning objectives of the tutorial#

  1. Discover the main challenges of DPM thanks to the estimation of Alzheimer’s disease progression at group and individual levels,

  2. Understand the rationale behind different state-of-the-art DPM methods and their limitations,

  3. Acquire operational knowledge for selecting a DPM suitable to any given dataset, and be able to implement it with the right software,

  4. Get an in-depth overview of the operational challenges of longitudinal data, along with the ‘know-hows’ to overcome them

Schedule#

[9:00 - 9:15] Introduction to Disease Progression Modelling

[9:15-11:35] Discrete DPM / by Neil Oxtoby & Vikram Venkatraghavan

Discrete models are capable of inferring a longitudinal picture of disease progression using only cross-sectional data. Europond : github link.

[11:45-13:00] Linear Mixed-effects models / by Igor Koval

Linear Mixed-effects Models are the workhorse of statistical analysis for longitudinal data. Important to understand their capabilities and limitations for analysing neurological disease progression.

[14:00-15:20] Parametric continuous DPM / by Igor Koval

Recent developments that address some limitations of Linear Mixed-effects methods account for the non-linear dynamics of progression. They are able to estimate a long-term history of changes based on longitudinal observations, referred to as a disease course mapping. Leaspy repository

[15:30-17:50] Non-parametric continuous DPM / by Sara Garbarino, Clément Abi Nader, Marco Lorenzi

DPM based on dynamical system modelling allow estimating the dynamics of biomarkers’ interactions characterizing the long-term disease history. Building upon probabilistic DPM (Lorenzi-NeuroImage-2017), these methods have been applied to the modelling of pathological protein propagation across brain networks in Alzheimer’s (Garbarino-IPMI-2019), and to simulate the effect of amyloid intervention on long term clinical and imaging outcomes (Abi Nader -preprint- 2020). GP progression repository.

[17:50-18:00] Wrap-up & Conclusion