POND2016

POND2016_logo

8 February, London

First International Workshop on Modelling the Progression Of Neurological Disease (POND2016)

Date: 8 February 2016
Location: Wellcome Collection, London, UK
Sponsor: Centre for Medical Image Computing (CMIC), University College London; in association with the EuroPOND consortium.

Progress on the challenge of spatio-temporal modelling of neurodegenerative disease has gained momentum rapidly through a series of recent publications. Such models have the potential to make major impact in the future management of neurological disease. This one-day event gathered together key international researchers on this emerging topic. It was a unique opportunity to share views and ideas on the latest methodological advances and to map the landscape of current and future approaches.


Program:

(click on a title to read the talk abstract, where available)

 

09.30 Registration

10.00 Opening: Danny Alexander (UCL)

10.10 Keynote speaker: Giovanni Frisoni (IRCCS Fatebenefratelli and University of Geneva)
"Marker discovery in neurodegenerative diseases"

Abstract: Multimodal disease modelling is not new in biomedicine. Knowledge on the dynamics of antibody and antigen changes in hepatitis B has been used since long for early and pre-symptomatic diagnosis, disease staging, follow-up of disease activity, and drug development. In the case of this viral disease, biomarker development is relatively easy thanks to the knowledge of the etiologic agent. In neurodegenerative diseases, etiology has been identified as the toxic build-up of specific proteinaceous aggregates (amyloid, tau, synuclein, TDP43). Hypothetical multimodal unidimensional dynamic models have been developed that recapitulate the main biological phenomena of Alzheimer’s disease (amyloidosis and tau-related neurodegeneration). These have been tested and shown to allow recognising the disease at an early pre-dementia and asymptomatic stage. Model building based on empirical data has been attempted, but limited to simple monodimensional piecewise time-limited models. The challenge of the Horizon2020-funded coSTREAM project is building a multimodal multidimensional dynamic model based on empirical data.

11.00 Break

11.20 Manifold learning of neurodegenerative processes

11.20 – 11.50 Ricardo Guerrero (Imperial College London) – "MR manifold learning in an elderly population as a neuro-imaging biomarker"

Abstract: Manifold learning refers to a set of machine learning techniques that aim at finding a low-dimensional representation of high-dimensional data while adequately representing the intrinsic geometry of the data. High dimensional datasets such brain MR images, are probably over-complete in the sense that a subspace of far fewer dimensions (that is most likely to be non-linear) may represent most of the variation in the data. Learning such a subspace representation remains a challenging task. In this talk I will discuss how manifold learning can be used to derive a low-dimensional representation from brain MR image intensities and how such subspaces can be used as to generate data-driven AD biomarkers. I will also show how this biomarkers can be used in modelling the longitudinal behavior of AD. 

11.50 – 12.20 Ashish Raj (Cornell University) – "Network models of neurodegeneration"

Abstract not available.

12.20 – 12.50 Yasser Iturria-Medina (McGill University) – "Towards Multi-factorial Data-Driven Modeling of Neurodegeneration"

Abstract: Late-onset Alzheimer’s disease (LOAD) is thought to be a multi-factorial disorder, but mechanisms underlying the disease’s progression are poorly characterized from an integrative perspective. We will present our recent work modeling spatiotemporal alterations in brain amyloid-β misfolded proteins, metabolism, vascular regulation, functional activity, structural properties, cognitive integrity, and periphery proteins levels in relation to LOAD progression. We analyzed over 5000 multi-modality brain images and tens of plasma and CSF biomarkers for healthy and diseased subjects from ADNI. Through a multi-factorial data-driven analysis, we studied the disease as a dynamic aging associated process, obtaining a tentative, data-driven, temporal ordering of disease progression. We will also highlight current limitations and challenges associated to the multi-factorial modeling of neurodegeneration.

12.50 Lunch break

13.50 Keynote speaker: Bruno Dubois (ICM – Hôpital Pitié Salpêtrière)
"Dementia modelling for treatment optimisation"

Abstract: The lack of success of trials using monoclonal antibodies targeting amyloid at a mild or moderate dementia stage of Alzheimer’s disease (AD) is a further encouragement to shift the attention to the earliest stages of the disease. This raises the question of the definition of AD in view of a treatment optimisation:
– Should it be defined as a clinical disease, which starts with a dementia?
This is the classical definition of the disease that is still learned.
– Should it be defined as a clinical disease, which starts with the first clinical symptoms, episodic memory impairment in most of the cases?
This is the definition that the International Working Group has proposed in 2007 considering that there is no reason to link the diagnosis of a disease to a certain threshold of severity. And this new definition has introduced the concept of the prodromal stage of the disease.
– Should it be defined biologically by the presence of a positive biomarker even in the absence of clinical symptoms?
According to this view, AD is a continuum and cognitively normal individuals, who are biomarker positive, already have the disease and will develop it further clinically.

Indeed, although we plead in favor of a definition of AD that integrates the earliest stages of the disease, even preclinical, several issues remain to be resolved. If in vivo evidence of Alzheimer pathology is a fundamental feature for the further progression to a clinical disease, it is not definitely established that all cognitively healthy subjects who are biomarker-positive will develop the disease during their lifetime. At present, the risk of conversion to a clinical AD has been estimated to be around 25% after a 3-year follow-up. Longer follow-ups are needed to demonstrate that all biomarker-positive cognitively normal subjects will progress to AD. In parallel, several factors may contribute to keeping the subjects under the threshold of clinical AD. We speculate that the occurrence of clinical onset of AD is the expression of a complex algorithm where the presence of AD brain lesions plays a central role, and other additional positive/negative factors need to be considered. For instance, several variables may have a negative effect such as age, presence of co-morbidity, vascular risk factors, genetic predisposition (ε4 allele in the apolipoprotein E, TOMM40,…), other unknown susceptibility genes and perhaps non-genetic risks. Other factors may compensate for all these aggressions: this may be the basis for a significant cognitive reserve or neuroplasticity, other biological compensatory mechanisms of the brain and preventive genetic/episodic and lifestyle factors that can delay or even prevent the occurrence of a clinical disease. For these reasons, cognitively normal biomarker-positive individuals should only be considered as “asymptomatic at risk for AD (AR-AD)”.

Therefore, the next steps will be to identify:
1) the factors that activate or delay the dynamic process of conversion; and
2) the subtle brain changes that anticipate the clinical onset.
We consider that the main clinical issue will be to identify AR-AD having the highest likelihood to convert to definite clinical AD in the few subsequent months. This is especially important, as the latter will be the main target population for treatment with disease-modifying drugs. When such drugs become available, the important step will be to treat AR-AD subjects to delay the onset of clinical symptoms.

14.40 EuroPOND Session

14.40 – 14.55 Jean-Baptiste Schiratti (ICM – Hôpital Pitié Salpêtrière) – "Learning spatiotemporal trajectories from manifold-valued longitudinal data"

Abstract: We will introduce a mixed-effects model to learn typical scenarios of changes from longitudinal manifold-valued data, namely repeated measurements of the same objects or individuals at several points in time. We will show that the model allows to estimate a group-average trajectory in the space of measurements and subject-specific trajectories of progression. The subject-specific trajectories result from spatiotemporal transformations of the average trajectory, where  the spatiotemporal transformations allow for changes in the direction of the trajectory and in the pace at which trajectories are followed.  A stochastic version of the Expectation-Maximization algorithm is used to estimate the model parameters in this highly non-linear setting. We will give experimental results in which the method is used to estimate a data-driven model of the progressive impairments of cognitive functions during the onset of Alzheimer’s disease. Results show that the model correctly put into correspondence the age at which each individual was diagnosed with the disease, thus validating the fact that it effectively estimated a normative scenario of disease progression. Random effects provide unique insights into the variations in the ordering and timing of the succession of cognitive impairments across different individuals.

14.55 – 15.10 Alexandra Young (UCL) –  "Event-based models of disease progression"

Abstract: The event-based model (Fonteijn et al., Neuroimage, 2012) is a probabilistic, data-driven model that can be used to determine the sequence in which biomarkers become abnormal from cross-sectional data, and to derive a fine-grained patient staging system, which can be used throughout the full disease time course. In this talk I will focus on two pieces of work. First, the application of the event-based model to the ADNI dataset (Young et al., Brain, 2014) to determine the sequence in which biomarkers become abnormal in sporadic Alzheimer’s disease, and the utility of the event-based model as a patient staging system. Second, the development of a new event-based model (Young et al., IPMI, 2015) that describes multiple sequences of events, allowing the determination of data-driven patient subgroups that follow distinct progression patterns.

15.10 Coffee Break

15.30 Temporal modelling of neurodegeneration from short-term clinical data

15.30 – 16.00 Bruno Jedynak (Johns Hopkins University / Portland State University) – "Modelling the progression of markers in aging and in Alzheimer’s disease"

Abstract: The Alzheimer’s Disease (AD) Neuroimaging Initiative (ADNI) was the first large multisite, longitudinal study designed to track the progression of AD using a wide range of markers assessing the brain’s structure and function over the course of the disease stages. It opened the possibility for computational scientists to contribute to the neurobiology of AD.  We review our earliest work on modelling the cascade of markers in ADNI assuming that all subjects in the study follow the same progression albeit with a different onset and different speed. This modelling approach allows for questioning the ordering of the markers in ADNI. We will then focus on two cohorts of healthy aging: the Baltimore Longitudinal Study of Aging (BLSA) and the Wisconsin Registry for Alzheimer’s Prevention (WRAP). We study a collection of eight cognitive tests spanning a large spectrum of cognitive domains. A refined version of our initial model, using mixed effects and estimating the covariance between the markers, allows for assigning to each subject a composite cognitive score. This score is computed by extrapolating to a standard age, e.g. 50 years old, the cognitive status of each individual. It provides a quantitative cognitive phenotype corrected for the age at which the subject is observed. We present simultaneous results on the WRAP and BLSA cohorts. Lastly, we present an alternative version of our model, allowing for a large collection of correlated markers.  Using data from the BLSA, we apply this model to the distribution volume ratio (DVR) images derived from Pittsburgh compound B (PiB) PET imaging, which show the distribution of cerebral fibrillar amyloid-beta (Aβ). We show that precuneus has the highest amyloid level among cortical regions at an early stage of amyloid accumulation.

16.00 – 16.30 Mike Donohue (University of Southern California) – "Mixing latent time and latent traits"

Abstract: We discuss regression approaches for estimating long-term multivariate progression (or growth) curves in the absence of long-term follow-up. We demonstrate an iterative backfitting algorithm and fully Bayesian approaches to simultaneous estimation of subject-specific (latent) “disease-time,” covariate effects, and long-term multivariate progression curves. We present basic simulations and preliminary analysis of Alzheimer’s Disease Neuroimaging Initiative (ADNI) data.

16.30 – 17.00 Gwenaëlle Douaud (University of Oxford) – 'From small sample size to Big Data, what can be gained from using different kind of datasets and advanced imaging analysis tools in Alzheimer's?'

Abstract: In this talk, I will present our most recent work in mild cognitive impairment and Alzheimer’s disease using structural imaging: starting with a small sample size prediction study of well characterised MCI patients using a lesser known DTI metric (Douaud et al., J Neurosci 2013), then a randomised clinical trial using Bayesian causal modelling (Douaud et al., PNAS 2013), and finally a multi-modal, data-driven approach on a big dataset of healthy subjects covering most of the lifespan (Douaud et al., PNAS 2014).

17.00 Closing remarks


Key publications relevant to POND2016:
  • Fonteijn, et al. An event-based model for disease progression and its application in familial Alzheimer’s disease and Huntington’s disease. NeuroImage, 60(3):1880-1889, 2012.
  • Jedynak, et al. A computational neurodegenerative disease progression score: Method and results with the Alzheimer’s Disease Neuroimaging Initiative cohort. NeuroImage, 63(3):1478–1486, 2012.
  • Donohue, et al. Estimating long-term multivariate progression from short-term data. Alzheimer’s & Dementia: The Journal of the Alzheimer’s Association, 10(5):S400–S410, 2014.
  • Young, et al. A data-driven model of biomarker changes in sporadic Alzheimer’s disease. Brain, 137(9):2564-2577, 2014.
  • Young, et al. Data-driven models of neurodegenerative disease. Advances in Clinical Neuroscience and Rehabilitation 14(5):6, 2014.
  • Huang and Alexander. Probabilistic Event Cascades for Alzheimer’s disease. In Advances in Neural Information Processing Systems 25, pages 3104-3112. 2012.
  • Bilgel, et al. Temporal Trajectory and Progression Score Estimation from Voxelwise Longitudinal Imaging Measures: Application to Amyloid Imaging. In 24th biennial international conference on Information Processing in Medical Imaging (IPMI 2015), Lecture Notes in Computer Science. Springer, 2015.
  • Lorenzi, et al. Efficient Gaussian Process-Based Modelling and Prediction of Image Time Series. In 24th biennial international conference on Information Processing in Medical Imaging (IPMI 2015), Lecture Notes in Computer Science. Springer, 2015.
  • Lorenzi, et al. Regional flux analysis for discovering and quantifying anatomical changes: An application to the brain morphometry in Alzheimer’s disease. NeuroImage,  115 (15): 224–234, 2015.
  • Lorenzi, et al. Disentangling Normal Aging from Alzheimer’s Disease in Structural MR Images. Neurobiology of Aging, 36 Suppl 1:S42-52, 2015.
  • Schmidt-Richberg, et al. Multi-stage Biomarker Models for Progression Estimation in Alzheimer’s Disease. In 24th biennial international conference on Information Processing in Medical Imaging (IPMI 2015), Lecture Notes in Computer Science. Springer, 2015.
  • Schiratti, et al. A mixed-effect model with time reparametrization for longitudinal univariate manifold-valued data. In 24th biennial international conference on Information Processing in Medical Imaging (IPMI 2015), Lecture Notes in Computer Science. Springer, 2015.
  • Young, et al. Multiple Orderings of Events in Disease Progression. In 24th biennial international conference on Information Processing in Medical Imaging (IPMI 2015), Lecture Notes in Computer Science. Springer, 2015.
  • Young, et al. A simulation system for biomarker evolution in neurodegenerative disease. Medical Image Analysis 26(1):47-56, 2015.
  • Iturria-Medina, et al. Epidemic Spreading Model to Characterize Misfolded Proteins Propagation in Aging and Associated Neurodegenerative Disorders. PLoS Comput Biol, 10(11):e1003956, 11 2014.
  • Raj, et al. A Network Diffusion Model of Disease Progression in Dementia. Neuron, 73(6):1204-1215, 2012.
  • Zhou, et al. Predicting Regional Neurodegeneration from the Healthy Brain Functional Connectome. Neuron, 73(6):1216–1227, 2012; Alzheimer’s & Dementia, 10(4):159, 2014.
  • Villemagne, et al. Amyloid β deposition, neurodegeneration, and cognitive decline in sporadic Alzheimer’s disease: a prospective cohort study. The Lancet Neurology, 12(4):357 – 367, 2013.
  • Oxtoby, et al. Learning Imaging Biomarker Trajectories from Noisy Alzheimer’s Disease Data Using a Bayesian Multilevel Model. In Bayesian and grAphical Models for Biomedical Imaging, Lecture Notes in Computer Science, 8677:85–94, 2014.
  • Durrleman, et al. Toward a comprehensive framework for the spatiotemporal statistical analysis of longitudinal shape data. International Journal of Computer Vision, 103(1):22-59, 2013.