Probabilistic Fitting of Active Shape Models

This project makes the code available which is used for active shape model (ASM) fitting in the publication
Probabilistic Fitting of Active Shape Models [1].

Getting Started

The Environment

We assume that you have sbt already installed. If not, please follow the instructions given here.

Building the Software

We recommend that you build the software as a JAR-file. To create the JAR-File, issue the command sbt assembly in the project root folder where also the build.sbt file is present. This should then generate the file target/scala-2.12/shape18-asm-sampling.jar

Alternatively, if you are familiar with sbt you can get started by simply running sbt in the project directory.

Preparing the Data

First you need to obtain the 20 training items of the SLIVER07 dataset [2]. The data set contains the ct-volumes and the segmentation maps. We provide our landmarks for those volumes. These are handclicked by a computer scientist, not by an internist. You should sort the volume data into the folders data/sliver/volume-ct and data/sliver/volume-labelmap like this:

data/sliver/
├── landmarks
│   ├── liver-orig001.json
│   ├── liver-orig002.json
│   └── ...
├── volume-ct
│   ├── liver-orig001.mhd
│   ├── liver-orig001.raw
│   ├── liver-orig002.mhd
│   └── ...
└── volume-labelmap
    ├── liver-seg001.mhd
    ├── liver-seg001.raw
    ├── liver-seg002.mhd
    └── ...

From the projects root directory, you can use the JAR-file to prepare and align the data using following command:

java -Xmx8g -cp target/scala-2.12/shape18-asm-sampling.jar probabilisticFittingASM.ImportData

The output will be written under the directory data/experiments.

As we are not allowed to distribute derived data from the SLIVER07 data (see the rules), we can not provide our registered meshes. So you have to use your registration to register the meshes from the directoy data/experiments/segmentation. The registered meshes should have the same format and filename and go into the directory data/experiments/registered.

Building the Models

Based on the registered data we build the ASM with the command:

java -Xmx8g -cp target/scala-2.12/shape18-asm-sampling.jar probabilisticFittingASM.BuildModels

If you want to run the leave-one-out experiment, issue the command a second time with additional option -l standing for leave-one-out. For the leave one out experiment, we allow the shape to deform more than the standard PCA-based model. For this we augment the model using the command:

java -Xmx8g -cp target/scala-2.12/shape18-asm-sampling.jar probabilisticFittingASM.AugmentModels

Fitting the Models

Having the models built, we can run the different fittings:

1. The standard fitting can be started with:

   java -Xmx8g -cp target/scala-2.12/shape18-asm-sampling.jar probabilisticFittingASM.BuildModels
   

   Add the -l option for using the leave-one-out model.

2. The fitting using sampling can be started with:

   java -Xmx8g -cp target/scala-2.12/shape18-asm-sampling.jar probabilisticFittingASM.BuildModels
   

   Add the -l option for using the leave-one-out model.

3. When you provide in addition some annotated lines, you can run the fitting using sampling incorporating the lines. The lines are provided as vtk mesh where only the vertex locations are used. Place your meshes in the folder data/experiments/lines having the same name as the meshes generated in data/experiments/segmentation. The command to fit with lines is:

   java -Xmx8g -cp target/scala-2.12/shape18-asm-sampling.jar probabilisticFittingASM.BuildModels
   

   Add the -l option for using the leave-one-out model.

Each experiments write the final segmentation as mesh into one of the following folders:

data/experiments/testInModel/standard
data/experiments/testInModel/sampling
data/experiments/testInModel/lines

data/experiments/leaveOneOut/standard
data/experiments/leaveOneOut/sampling
data/experiments/leaveOneOut/lines

Evaluating the Results

To evalute the segmentations run the following script:

java -Xmx8g -cp target/scala-2.12/shape18-asm-sampling.jar probabilisticFittingASM.BuildModels

This command will fill the following two folder with files reporting the evaluation measures and initial plots:

data/experiments/testInModel/statistics

data/experiments/leaveOneOut/statistics

Making the Plots

The plots of the paper can be produced with the help of R. The script for that is located in resources/scripts.

Learning more ...

When you are interested to learn more about our software and the theory behind it, please have a look at our GPMM website with tutorials and online courses.

References

[1]	A. Morel-Forster, Th. Gerig, M. Lüthi, Th. Vetter: Probabilistic Fitting of Actiave Shape Models. ShapeMi Workshop, MICCAI (2018)
[2]	Heimann, T., van Ginneken et al., B.: Comparison and evaluation of methods for liver segmentation from ct datasets. IEEE Transactions on Medical Imaging 28(8), 1251–1265 (Aug 2009) https://doi.org/10.1109/TMI.2009.2013851