Export slicer DB as nifti

vglrun /opt/apps/slicer/Slicer-4.10.2-linux-amd64/Slicer --no-main-window --python-script ./anonymize.py

longitudinal registration

make -Bki -j10 labelbcm make -Bki -j10 maskbcm make -Bki -j8 maurerbcm make -ki -j 8 longregbcm make -ki -j 8 distregbcm

epmstats

make -k -i -B -f lrstatistics.makefile epm make -k -i -B -f lrstatistics.makefile epmstatdf cat epmstats/*/lstat.csv > epmstats/lstat.csv cat epmstatistics.sql | sqlite3 Rscript epmstats.R

autoepmstats

make -k -i -B -f benignlrstats.makefile epmauto make -k -i -f benignlrstats.makefile epmstatauto cat autostats/*/lstat.csv > autostats/lstat.csv cat epmautomanual.sql | sqlite3 Rscript automanualepm.Rmd

benign epmstats

make -k -i -B -f benignlrstats.makefile fixedtrain make -k -i -B -f benignlrstats.makefile epmbenign make -k -i -B -f benignlrstats.makefile epmstatbenign cat epmstats/*/lstat.csv > epmstats/lstat.csv cat epmautostats.sql | sqlite3 Rscript autoepmstats.R

Matlab example

https://www.mathworks.com/help/images/segment-3d-brain-tumor-using-deep-learning.html

litstumor model

make -f lits.makefile crctumor make -f lits.makefile analyze make -f lits.makefile preprocess make -f lits.makefile train

Usage

convert dicome to nifti and anonymize data

make art truth

make anon

preprocess - intensity normalize

make scaled

preprocess - resize - Data set with a valid size for 3-D U-Net is multiple of 8

make resize

preprocess - python code is used to setup the kfold fold

preprocess - each fold is configured with a json file to be read by matlab

python setupmodel.py --databaseid=hccmri --initialize python setupmodel.py --databaseid=hccmri --setuptestset

train models

matlab livermodel.m

apply NN to test set

make mask

evaluate accuracy

make overlap

detection

make -ki -f methodist.makefile -j 12 rawmethodist biasmth make -ki -f methodist.makefile labelmth make -ki -f methodist.makefile -j 12 maskmth make -ki -f methodist.makefile -j 12 longregmth

matlab code structure

ImageSegmentationBaseClass.m - ABC defining the interface ImageSegmentationDeepMedic.m - derived class for deep medic architecture ImageSegmentationDensenet2D.m - derived class for Densenet2D architecture ImageSegmentationDensenet3D.m - derived class for Densenet3D architecture ImageSegmentationUnet2D.m - derived class for Unet2D architecture ImageSegmentationUnet3D.m - derived class for Unet3D architecture

hccmriunet3d.m - derived class using the 3d unet to segment hcc on mri hccmriunet2d.m - derived class using the 2d unet to segment hcc on mri

TF Usage

python lrmodel.py --databaseid=lrbcm --initialize python lrmodel.py --databaseid=lrbcm --setuptestset python lrmodel.py --databaseid=lrbcm --builddb make -f lrbcm256kfold010.makefile models tensorboard --logdir=./lrbcmlog/ --port=6010 make -f lrstatistics.makefile labels make -j 8 -B -f lrstatistics.makefile lstat cat qastats/*/lstat.csv > qastats/lstat.csv cat lrstatistics.sql | sqlite3 R ; source('lrstats.R')