3D/2D vessel/catheter-based registration and 3D catheter tip tracking using hidden Markov model

This code is the implementation of the methods presented in the following papers:

P. Ambrosini, I. Smal, D. Ruijters, W. J. Niessen, A. Moelker and T. van Walsum: A Hidden Markov Model for 3D Catheter Tip Tracking With 2D X-ray Catheterization Sequence and 3D Rotational Angiography. IEEE Transactions on Medical Imaging, vol. 36(3), pp. 757-768, 2017.
P. Ambrosini, D. Ruijters, W.J. Niessen, A. Moelker and T. van Walsum: Continuous Roadmapping in Liver TACE Procedures Using 2D-3D Catheter-based Registration. International Journal of Computer Assisted Radiology and Surgery, vol. 10, pp. 1357-1370, 2015.

Prerequisites

Tested on:

Windows 10 64bits with:
- Microsoft Visual C++ 14.00 2015 update 3 express version
- Microsoft Visual C++ 12.00 2013 update 5 express version
- MinGW-w64 and GCC 5.4.0

It should work on Unix-like systems as well with small changes (not tested). Note that multithreading and AVX instructions are not fully supported.

To build the code you need:

CMake to build the project (tested with 3.9.0-rc5)
ITK library mainly for the Powell optimizer and ITK image routines (tested with 4.12.0)

Quick start

On Windows, execute these commands to build the executable. ITK_DIR is the path where your ITK library is.

cd cpp
mkdir generated
cd generated
cmake "../" -DCMAKE_CONFIGURATION_TYPES="Debug;Release" -DUSE_ITK=ON -DITK_DIR="C:/libs/itk"
cmake --build . --config Release

Then, run the example with the configuration proposed in [1] (i.e. 3D catheter tip tracking using hidden Markov model). ../../data/dataset1 is the path where the dataset is (3D vessels centerline file, initial 3D catheter tip position, 2D catheter centerline files, C-arm angle and position). configHMMPowell.txt is the configuration file (more information in the Specifications section). generated is the path where to save the results. -debugImages says that we want to have debug images to check the registrations.

cd ../../examples/HMMPowell
mkdir generated
"../../cpp/generated/Release/TACE" Registration "../../data/dataset1" "configHMMPowell.txt" "generated" -debugImages

In the folder generated you should have now the transformation matrices and images of the registration between the 2D catheter (in red) and the 3D vessels (in green) for the sequence of 74 images.

Specific build

In order to have your own building configuration with none default parameters, we propose a different way (actually the standard way) to build the project. First edit these two files userSpecific/globalVariables.default.bat and userSpecific/buildConfig.default.bat and change all the paths/parameters to fit your machine configuration. Then, build the project.

cd cpp
buildConfigRedirect.bat
compileRedirect.bat

To see if everything went fine, check the files cpp/generated/stderrbuildConfig.txt, cpp/generated/stdoutbuildConfig.txt, cpp/generated/stderrcompile.txt and cpp/generated/stdoutcompile.txt.

Finally examples are simply launched with a batch script.

cd ../examples/HMMPowell
HMMPowellRedirect.bat

Output can be checked in the files generated/stderrHMMPowell.txt and generated/stdoutHMMPowell.txt.

Examples

examples/HMMPowell is the method with the optimal parameters described in [1]. You can play with the parameters in configHMMPowell.txt.
examples/HMMBruteforce is the same as the previous example but with brute force optimizer instead of Powell optimizer. It is much slower but should be more likely to stop to the global optimum.
examples/shapeSimilarityBruteforce is the method with the optimal parameters and brute force optimizer described in [2].
examples/shapeSimilarityPowell is the brute force method with the optimal parameters and Powell optimizer described in [2].

Specifications

Configuration file

The configuration file in the examples has the following parameters:

m_Method (METHOD_SHAPE_SIMILARITY = 0, METHOD_HMM = 1)
- METHOD_SHAPE_SIMILARITY: 2D/3D registration method using shape similarity [2].
- METHOD_HMM: 2D/3D registration method using hidden Markov model [1].
m_Optimizer (OPTIMIZER_BRUTE_FORCE = 0, OPTIMIZER_POWELL = 1)
- OPTIMIZER_BRUTE_FORCE: It is an exhaustive search, detail of the method is described in [2] (slow, more accurate).
- OPTIMIZER_POWELL: It is the Powell optimizer (fast, less accurate if not initialized carefully).
m_DistanceUsed (DISTANCE_SQR = 0, DISTANCE_ABSOLUTE = 1, DISTANCE_EUCLIDEAN = 2): Specify how is compute the minimum distance between points.
- DISTANCE_SQR: Square distance (fast).
- DISTANCE_ABSOLUTE: Absolute distance (fastest).
- DISTANCE_EUCLIDEAN: Euclidean distance (slowest).
m_MetricUsed (METRIC_FROM_CATHETER = 0, METRIC_IMPROVED_FROM_CATHETER = 1): Specify how is compute the minimum distance metric.
- METRIC_FROM_CATHETER: Minimum distance to all points, similar to eq.4 D1(lsel, T) [2].
- METRIC_IMPROVED_FROM_CATHETER: Minimum distance in a certain range along the path, eq.5 D(ci, lsel, T, pprev) [2].
m_Dmax // in mm (between 10 and 150mm), use only if m_MetricUsed = METRIC_IMPROVED_FROM_CATHETER: It is the neighborhoud range for the minimum distance metric D, called dmax [2].
m_UseWeight (True, False): Use or do not use a weight function to give more weight close to the tip in the minimum distance metric, eq.7 W(x) [2].
m_WeightSigma // sigma (between 20 and 100 with a catheter size in average of 200mm) of the gaussian W(x) = lambda + (1.0 - lambda)e(-x^2/(2sigma^2)), use only if m_UseWeight = True: Sigma in the weight function W(x) [2].
m_WeightLambda // lambda (something between 0 and 0.5), use only if m_UseWeight = True: Lambda in the weight function W(x) [2].
m_RadiusMetric (RADIUS_METRIC_FCOST1 = 0, RADIUS_METRIC_FCOST2 = 1, RADIUS_METRIC_FCOST3 = 2): Take or do not take into account the radius in the minimum distance metric.
- RADIUS_METRIC_FCOST1: Do not take into account the radius. Function Fcost1(c, Vi, tau) [1].
- RADIUS_METRIC_FCOST2: Take into account the radius. Give with less penalty when a 2D point closest is in the radius of a 2D projected 3D point. Function Fcost2(c, Vi, tau) [1].
- RADIUS_METRIC_FCOST3: Undocumented.
m_RadiusAlpha1, use only if m_RadiusMetric = RADIUS_METRIC_FCOST2 or RADIUS_METRIC_FCOST3: Alpha1 in Fcost2(c, Vi, tau) [1].
m_RadiusAlpha2, use only if m_RadiusMetric = RADIUS_METRIC_FCOST2 or RADIUS_METRIC_FCOST3: Alpha2 in Fcost2(c, Vi, tau) [1].
m_NumberStateEvaluatedNo: No [1].
m_SigmaS: sigma_s [1]
m_TransitionMetric (TRANSITION_METRIC_GAUSSIAN_A_PRIME = 0, TRANSITION_METRIC_BINARY_A_DOUBLE_PRIME = 1)
- TRANSITION_METRIC_GAUSSIAN_A_PRIME: a' [1]
- TRANSITION_METRIC_BINARY_A_DOUBLE_PRIME: a'' [1]
m_SigmaA // in mm, used only if m_TransitionMetric = TRANSITION_METRIC_GAUSSIAN_A_PRIME: sigma_a [1]
m_Theta // in mm, used only if m_TransitionMetric = TRANSITION_METRIC_BINARY_A_DOUBLE_PRIME: theta [1]

3D vessel centerline structure

The structure of the 3D vessel is text file with the following lines:

number of points
point id of the vessel root (usually it's always 0, i.e. the first point)
first point position, direction, radius and orthogonal direction (direction and radius are a rough estimation) separated by a ';' (Format is "x; y; z; dir_x; dir_y; dir_z; radius; ortho_dir_x; ortho_dir_y; ortho_dir_z; 0; 0; 0; 0; 0; branch_id")
point id neighbors of the first point separated by a ';'
second point position, direction, radius and orthogonal direction (direction and radius are a rough estimation) separated by a ';'
point id neighbors of the second point separated by a ';'
...
last point position, direction, radius and orthogonal direction (direction and radius are a rough estimation) separated by a ';'
point id neighbors of the last point separated by a ';'

Example: Vessel with three points. The first point is the root and is linked to the two other points.

3
0
5.0;4.0;4.0;0.;0.;1.;0.5;1.;1.;0.;0.;0.;0.;0.;0.;0
1;2;
5.0;4.0;5.0;0.;0.;1.;0.5;1.;1.;0.;0.;0.;0.;0.;0.;1
0;
5.0;3.0;3.0;0.;0.707;0.707;0.5;0.;0.;1.;0.;0.;0.;0.;0.;2
0;

HaohuiHU/HMM-3D-Catheter-Tip-Tracking