Simulating Neurodegeneration through Longitudinal Population - - PowerPoint PPT Presentation

Outline Background Simulating neurodegneration through longitudinal population analysis

Simulating Neurodegeneration through Longitudinal Population Analysis of Structural and Diffusion Weighted MRI Data

Modat et al. MICCAI 2014 Bishesh Khanal, Computer Vision Seminar

Asclepios, INRIA Sophia Antipolis

October 8, 2014

1 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

1

Background Brain and Neurodegeration Alzheimer’s Disease and Multimodal NeuroImaging Atrophy Measurement and Registration

2

Simulating neurodegneration through longitudinal population analysis Overall Framework Multimodal Registration and Template database Simulating Flows Discussion

2 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

1

Background Brain and Neurodegeration Alzheimer’s Disease and Multimodal NeuroImaging Atrophy Measurement and Registration

2

Simulating neurodegneration through longitudinal population analysis Overall Framework Multimodal Registration and Template database Simulating Flows Discussion

3 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Neurons, Gray Matter(GM), White Matter(WM) and Cerbral Spinal Fluid (CSF)

Figure:

Cortex.[http://en.wikipedia.org/wiki/File: Neuron_Hand-tuned.svg. Accessed on 08-01-2014]

Figure:

GM/WM.[http://www.medinewsdigest.com/?p=3249. Accessed on 08-01-2014] 4 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Neurodegeneration and Longitudinal images

Figure: Baseline and follow-up images of an Alzheimer’s Disease patient.

Neurodegenration: Progressive loss of structure or function

• f neurons, including death of neurons [wikipedia]

Longitudinal images: Time series images of a same subject.

5 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

1

Background Brain and Neurodegeration Alzheimer’s Disease and Multimodal NeuroImaging Atrophy Measurement and Registration

2

Simulating neurodegneration through longitudinal population analysis Overall Framework Multimodal Registration and Template database Simulating Flows Discussion

6 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Dementia and Alzheimer’s Disease

Dementia: Progressive decline of cognitive functions.

Loss of memory, mood changes, and problem with communication and reasoning.

AD: Most common cause of Dementia, mostly affects older people.

Characterized by atrophy, Amyloid β (Aβ) plaques and Neurofibrillary tangles (NFTs). Figure: Aβ plaques and NFTs in AD [2000 BrightFocus Foundation]

7 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Multimodal neuroimaging for AD

Different modalities required to get different types of information.

Figure: Amyloid imaging and structural MRI Figure: Tractography example

[http://en.wikipedia.org/wiki/File: DTI-sagittal-fibers.jpg] 8 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

1

Background Brain and Neurodegeration Alzheimer’s Disease and Multimodal NeuroImaging Atrophy Measurement and Registration

2

Simulating neurodegneration through longitudinal population analysis Overall Framework Multimodal Registration and Template database Simulating Flows Discussion

9 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Atrophy Measurement

Figure: Baseline and follow-up images of an Alzheimer’s Disease patient.

Segmentation based methods. Registration based methods. How to validate these methods ?

10 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Registration

Figure: Nonlinear registeration of longitudinal images

11 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Registration

Figure: Example displacement fields of longitudinal evolution

12 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Registration

Figure: Registration overview [ Sotiras et al BIASS2013 ]

13 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

1

Background Brain and Neurodegeration Alzheimer’s Disease and Multimodal NeuroImaging Atrophy Measurement and Registration

2

Simulating neurodegneration through longitudinal population analysis Overall Framework Multimodal Registration and Template database Simulating Flows Discussion

14 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Figure: Framework of the overall pipeline

15 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

1

Background Brain and Neurodegeration Alzheimer’s Disease and Multimodal NeuroImaging Atrophy Measurement and Registration

2

Simulating neurodegneration through longitudinal population analysis Overall Framework Multimodal Registration and Template database Simulating Flows Discussion

16 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Weighting similarity measures from different modalities

Similarity measure M that drives the registration: M(B, F; µ) = α × Ms(Bs, F s(u; µ)) + α × Ms(Bs(u−1; µ), F s) +β × Md(Bd, F d(u; µ)) + β × Md(Bd(u−1; µ), F d) (1) where,

s: Structural T1 image d: Diffusion weighted image

B: Baseline image F: Followup image Ms: locally normalised cross correlation summed over all voxels. Md: distance between the tensors, summed over all voxels. α, β: Weights (empirically set to 0.5) u: Deformation field µ: parameters, the cubic b-spline parameters.

17 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Template database

Figure: Template database

18 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

1

Background Brain and Neurodegeration Alzheimer’s Disease and Multimodal NeuroImaging Atrophy Measurement and Registration

2

Simulating neurodegneration through longitudinal population analysis Overall Framework Multimodal Registration and Template database Simulating Flows Discussion

19 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Registering new subject

For a new subject i with its T1w and DTI image Bi, Register all baseline images from the template database to Bi.

Figure: Schematic for registrations between template and the subject

20 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Weighting the flows to create a new one

Weight flows based on the distance of the template’s basline image to subject image distance. vgrp

i

=

• j∈grp
• ui

j ◦ vj

• × e−D

(Bi ,Bj ;ui j ) t

• j∈grp e−D

(Bi ,Bj ;ui j ) t

(2) Simulated follow-up images F NC

i

, F FTD

i

and F AD

i

simulated from Bi as: F grp

i

= exp(vgrp

i

) ◦ Bi

21 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Simulation Example

Figure: Subject and disease-specific longtudinal changes simulator result

22 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Simulation Results

Figure: Distance sorted in ascending order

23 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Simulation Results

Figure: Distance sorted in ascending order

24 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Discussion

Using multi-modal data decreases variance of registration

• results. (Needs further experimental verification).

Flow propagation method: Parallel transport and other techniques ? Can we extend to ”learning of deformation fields” ?

25 / 26

Outline Background Simulating neurodegneration through longitudinal population analysis

Thankyou!

26 / 26