README.md

LV segmentation

Authors:

Alberto Gomez ([email protected]) Miguel Xochicale ([email protected]) Nhat Phung ([email protected])

Summary

This plug-in does a segmentation of the LV in ultrasound images. The model is implemented in Pytorch.

Usage

After building the standalone software PRETUS, and after adding the path where this plug-in is installed in the pretus config file (~/.config/iFIND/PRETUS.conf), you should see the plug-in and it's help when running pretus:

...
(15) Plugin Name: 'LV Seg'

# PLUGIN LV Seg
   Automatic Left Ventricle segmentation using Unet
	--lvseg_stream <val> [ type: STRING]	Name of the stream(s) that this plug-in takes as input. (Default: ) 
	--lvseg_layer <val> [ type: INT]	Number of the input layer to pass to the processing task. If negative, starts 
                                 		from te end. (Default: 0) 
	--lvseg_framerate <val> [ type: FLOAT]	Frame rate at which the plugin does the work. (Default: 20) 
	--lvseg_verbose <val> [ type: BOOL]	Whether to print debug information (1) or not (0). (Default: 0) 
	--lvseg_time <val> [ type: BOOL]	Whether to measure execution time (1) or not (0). (Default: 0) 
	--lvseg_showimage <val> [ type: INT]	Whether to display realtime image outputs in the central window (1) or not (0). 
                                     		(Default: <1 for input plugins, 0 for the rest>) 
	--lvseg_showwidget <val> [ type: INT]	Whether to display widget with plugin information (1-4) or not (0). Location is 
                                      		1- top left, 2- top right, 3-bottom left, 4-bottom right. (Default: visible, 
                                      		default location depends on widget.) 
   Plugin-specific arguments:
	--lvseg_modelname <*.pth> [ type: STRING]	Model file name (without folder). (Default: model.pth) 
	--lvseg_cropbounds xmin:ymin:width:height [ type: STRING]	set of four colon-delimited numbers with the pixels to define the crop bounds 
                                                          		(Default: 0.25:0.2:0.55:0.65) 
	--lvseg_abscropbounds 0/1 [ type: BOOL]	whether the crop bounds are provided in relative values (0 - in %) or absolute 
                                        		(1 -in pixels) (Default: 0) 

Build instructions

Dependencies

The minimum requirements are:

• VTK. You need to fill in the VTK
• ITK (for video inputs, built with the ITKVideoBridgeOpencv option ON). You need to fill in the VTK
• Boost
• Qt 5 (tested with 5.12). You need to fill in the QT_DIR variable in CMake
• c++11
• Python (you need the python libraries, include dirs)

Additionally, for this plug-in:

• Python 3 (tested on Python 3.7)
• Python 3 library
• PyBind11 (for the python interface if required), with python 3.
• matplotlib
• numpy
• opencv-python
• scikit-learn
• scipy
• torch 1.14.0

The python include and binary should be the same used for pybind11. For example, if the python distribution comes from Anaconda, your PYTHON_INCLUDE_DIR in the CMake will be something like <HOME_FOLDER>/anaconda3/include/python3.7m and your PYTHON_LIBRARY will be something like <HOME_FOLDER>/anaconda3/lib/libpython3.7m.so.

Build and install

Launch CMake configure and generate. You will need to fill in the following (PLease ignore the references to MUSCLE and consider LV instead):

• QT:

• ITK/VTK:

• Plugin lib (from PRETUS):

• Python stuff (make sure this matches your conda environment for pretus):

• Also change your CMAKE_INSTALL_PREFIX to where you want to install the plug-in, typically <HOME>/local/VITAL-muscle.

Then make and install and launch.