Simulation of retinal hemodynamics

This code corresponds to our MICCAI 2018 paper on retinal hemodynamics simulation. If you use this code, please cite: Orlando, JI, Barbosa Breda, J, van Keer, K, Blaschko, MB, Blanco, PJ and Bulant, C. "Towards a glaucoma risk index based on simulated hemodynamics from fundus images". MICCAI 2018.

Installation

Clone the repository doing git clone https://github.com/ignaciorlando/retinal-hemodynamics.git.
Run git submodule update --recursive
Move to external/cnn-finetune/dependencies and run git submodule update --init
Open MATLAB and move to the root folder of the repository.
Run setup in the MATLAB console to add the folder to the path.

--- Note --- If you get an error when compiling Matconvnet, it might be that you don't have libdevjpeg installed in your computer. Run sudo apt-get install libjpeg-dev.

--- Note --- Once the project is setup, you just need to call the script setup_add_folders to automatically add the project folders to the MATLAB path.

Data preparation

We used some publicly available data sets for our experiments. In order to set up the data in a proper way, you can use some of the scripts in the data-organization folder.

script_setup_rite_data: downloads the RITE dataset and generates folders for arteries and veins.
script_setup_origa650_data: preprocess the images in the ORIGA650 database for fine-tuning the CNN.

We have also introduced a new database, namely Leuven Eye Study, comprising 22 fundus pictures with vessel annotations and labels for arteries and veins. We are working on the release right now!

script_setup_leuven_eye_study: organizes the data of the Leuven Eye Study for our experiments.

Precomputed resources

The folder ./precomputed-data contains precomputed data such as manual optic disc segmentations.

Data structures

Arterial branches are mapped to an intermediate graph representation, which is obtained from the centerline of the arterial segmentation. This graph structure is a MATLAB struct with the following fields:

node: is a cell array of nodes. Each node has the following fields:
- node.idx: array of indices of pixels in the branching/root node.
- node.links: array with the ids of the links spanning from the node.
- node.conn: array with the ids of the nodes connected with the current node through a link. The order of the ids coincides with the order of the links in node.links.
- node.numLinks: length of the array of links.
- node.is_root: a logical value indicating if the node is a root.
- node.tree_id: id of the tree where the node is located.
- node.comx: mean x coordinate of the node.
- node.comy: mean y coordinate of the node.
link: is a cell array of links.
- link.n1: one of the nodes connected by the link.
- link.n2: the other node connected by the link. If the link is a terminal link, n2 is equal to -1.
- link.point: list of points in the centerline that are part of the link.
- link.tree_id: id of the tree where the link is located.
w: is the width of the associated image.
h: is the hight of the associated image.
roots: is an array with the ids of the root nodes for each subtree. The position in the array is associated with the tree_id, while the content of the position is the id of the node which is root of that tree.

Experiments

Manual labeling

script_delineate_od

The root nodes of the computation graph are determined by analyzing the intersection between the arterial trees and the optic disc. To this end, it is essential to count with a (corse) segmentation of the optic disc. This script allows you to manually delineate an ellipse around the optic disc, and saves the output segmentations in a separate folder.

Input data generation

script_generate_input_data

This script allows you to compute the input data needed to perform the hemodynamic simulation and saves it as a .MAT file. In particular, it computes the arterial trees centerlines (trees_ids), their radius at each point (trees_radius) and the graph structure (graph).

--- Important! --- The data set that you will process will require to count with the arterial segmentations (in an arteries folder) and the optic disc segmentations (in an od-masks).

Figure generation

display_graph(graph, [image])

Display a given graph. If image is given, the graph is display over the input image.

VTK data generation

script_generate_input_data_vtk

This script allows you to export the skeletonizations generated with the script_generate_input_data script to VTK files. Such files are the imput to the code that computes hemodynamics. The pixel spacing is hardcoded in this script.

Run hemodynamic simulations

script_run_simulation

This script allows you to run a set of scenarios defined by a total blood flow and ophthalmic pressures. It is mandatory to generate the VTK files before running this script. The pixel spacing is hardcoded in this script, and should match the one used to generate the VTK files. The scripts reads data folders from config_generate_input_data. The output is stored in the folder RITE-<test|training>/hemodynamic-simulation/.

For each scenario, a .VTK and a .mat files contaning the simulation results are generated. The file name follows the encoding: <ImageID>_<test|training>_SC<scenarioID>_sol.<mat|vtk>.

The .mat file contains a variable named sol, which is a 3D matrix of dimensions [width,height,8]. Matrix dimensions represents (1) the radius (in [cm]), (2) the flow (in [ml/s]), (3) the pressure (in [mmHg]), (4) the velocity (in [cm/s]), (5) resistance, (6) Reynolds number, (7) wall shear stress, and (8) a mask indicating if the centerline pixel is arterial segment (0), a root (1), a terminal (2) or a bifurcation (3). All the files have NaN values on non-centerline pixels.

Export simulation results to MATLAB format

script_export_vtksimsol_matlab

Use this script to export the results of the simulations to a MATLAB-friendly format. This will also prepare the data for running the experiments using the BoHF.

Run classification experiment using BoHF

script_bohf_cross_validation

This script runs the classification experiment using the BoHF. You need to modify the configuration file config_bohf_cross_validation accordingly.

Perform data analysis

script_hemodynamics_sensitivity_analysis

This script allows you to exploratory data analysis and statistics on the hemodynamics simulations results. UNDER DEVELOPMENT!

desti-nation/retinal-hemodynamics