Atlas guided EM Algorithm

This implementation is based on Shimizu, A., Ohno, R., Ikegami, T. et al. (2007) Segmentation of multiple organs in non-contrast 3D abdominal CT images, Int J CARS 2: 135. https://doi.org/10.1007/s11548-007-0135-z.

Note this is not official implementation.

Please use at your own responsibility.

Requirements

Nuget package

• Eigen

• Boost

• ITK

Another library (already contained)

• mist

Environment

• Windows 10 pro

• Visual Studio 2015

• Release x64

Detail of data

Data I used is as follow:

Label correspondense

Organs	Label number
others	0
liver	1
spleen	2
kidneyL	3
kidneyR	4
heart	5
gallbladder	6
pancreas	7
aorta	8
IVC	9
portal	10
stomachWall	11
oesophagus	12

Note: In this case, i didnt use heart label since it is not included in mask image region.

Example:

Original	Label	Mask

How to use

First of all, please build all project.

And then, please put your data into data/preprocessed folder.

data
└─preprocessed
   ├─abdominal_cavity_mask        <--- mask image e.g. region of interest image
   ├─label                        <--- label image
   └─org                          <--- original image

I show flow chart as bellow:

Cite

1. generate_atlas

Generate atlas.

This program outputs into results/generate_atlas.

bat\generate_atlas.bat

2. compute_features

Compute features each original image.

In this case, we calculate median image.

bat\compute_features.bat

3. calc_init_val

Compute initial value for em algorithm using training data.

bat\calc_init_val.bat

4. apply_em_algorithm

Apply em algorithm to test data

bat\apply_em_algorithm_using_atlas.bat

5. apply_maximum_a_posteriori

Apply maximum a posteriori to test data

bat\apply_maximum_a_posteriori.bat

Results

I show example of result.

Predict	Ground truth	Original

I evaluate my result in terms of Jaccard Index.

organs	JI
liver	0.862857
spleen	0.616608
kidneyL	0.666968
kidneyR	0.548733
gallbladder	0.187473
pancreas	0.35192
aorta	0.606638
IVC	0.566764
portal	0.0789795
stomachWall	0.13219
oesophagus	0.0801887
other	0.728133