Data-Driven Inference of Vertexwise Evolution
Switch branches/tags
Nothing to show
Clone or download
Pull request Compare This branch is even with mrazvan22:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
DEM.py
DisProgBuilder.py
Makefile
PlotterVDPM.py
README.md
VDPMLinear.py
VDPMMean.py
VDPMNan.py
VDPMNanMasks.py
VDPMNanNonMean.py
VDPMSimplified.py
VDPM_MRF.py
adniCommon.py
adniDEM.py
adniPet.py
adniThick.py
aligners.py
aux.py
blenderCol.py
colorAtrophyExtent.py
colorClustProb.py
colorClustSlope.py
colorDiffs.py
colorVerticesDirectly.py
diffEqModel.py
env.py
evaluationFramework.py
launchCommon.py
launchSynth.py
launchThick.py
plotFunc.py
procADNICluster.py
procADNIMerge.py
procADNIPET.py
procPETMerge.py
synth.py
synthCommon.py
synthMRF.py
tadpole.py
voxCommon.py
voxelDPM.py

README.md

== Getting started ==

=== Synthetic tests ===

Run the DIVE model on several synthetic tests using the following command:

make synthAll

This will run three main synthetic scenarios:

  1. cluster trajectories getting closer to one another
  2. increasing number of clusters
  3. decreasing number of subjects

Each scenarioruns 8 successive steps that get harder and harder (e.g. many subjects generated -> very few subjects generated). The performance of DIVE is expected to decrease as the scenarios increase in difficulty.

=== Real data ===

In order to get DIVE working on real data, you need to use some images where the voxels are aligned across subjects and timepoints. You can use any software you prefer for the image registration (I used Freesurfer). When applying to a different dataset, DIVE might require setting suitable priors and initialisations of the parameters.

== Implementation details ==

The main class of the model is in voxelDPM.py. The launcher scripts used to start various experiments are launcher***.py, and the VDPM***.py are different flavours of the model, which can use linear trajectories, allow missing data or perform spatial regularisation with a Markov Random Field. I recommend you use the VDPMMean (model index 8), because it uses a mathematical optimisation that yields a x20 decrease in computation time. I am currently finishing the journal paper describing all of these.

== References ==

[1] A vertex clustering model for disease progression: Application to cortical thickness images, Razvan V. Marinescu, Arman Eshaghi, Marco Lorenzi, Alexandra L. Young, Neil P. Oxtoby, Sara Garbarino, Timothy J. Shakespeare, Sebastian J. Crutch and Daniel C. Alexander http://www.homepages.ucl.ac.uk/~rmaprvm/MarinescuIPMI2017.pdf