pl-lld_inference

Table of Contents

Abstract
Description
Usage
- Arguments
- Run
Development
Examples

Abstract

This plugin application is part of the Leg-Length-Discrepancy project that measures leg lengths based off X-Ray images. This particular plugin operates on input mha format images, and infers the locations of several anatomical landmarks on the leg images. These locations are shown as heat maps (i.e. small fuzzy regions of probalities) that are superimposed on the original images. In addition, a prediction text file in csv format with the locations of the heatmaps is also generated.

Description

lld_inference is a ChRIS ds-type application that works off inputs of leg images (X-Ray images converted to *mha format). For each image, the plugin attempts to determine where six landmarks are located -- three in each leg (in layman's terms):

top (or superior) of femur bone (close to the hip around the intertrochanteric line)
middle of the knee, slightly inferior to the intercondylar notch
bottom (or inferior) of the tibial bone (just inferior to the medial malleolus)

For each inferred location of these three landmark in each leg, the plugin generates a heatmap image for that landmark. Thus, six separate images are created -- blank everywhere except for the heatmap. Taken all together, these six images are then superimposed onto the original image for comparison. In addition to output images, the plugin also generates a text file in csv format that describes the heatmap centroids in (x, y) coordinate pairs.

Usage

docker run --rm fnndsc/pl-lld_inference lld_inference
    [-f|--inputFileFilter <inputFileFilter>]
    [-h|--help]
    [--json] [--man] [--meta]
    [--savejson <DIR>]
    [-v|--verbosity <level>]
    [--version]
    <inputDir> <outputDir>

Arguments

[-f|--inputFileFilter <inputFileFilter>]
A glob pattern string, default is "**/*.mha", representing the input
file pattern to analyze.

[-h] [--help]
If specified, show help message and exit.

[--json]
If specified, show json representation of app and exit.

[--man]
If specified, print (this) man page and exit.

[--meta]
If specified, print plugin meta data and exit.

[--savejson <DIR>]
If specified, save json representation file to DIR and exit.

[-v <level>] [--verbosity <level>]
Verbosity level for app. Not used currently.

[--version]
If specified, print version number and exit.

Getting inline help is:

docker run --rm fnndsc/pl-lld_inference lld_inference --man

Run

Attention!

pl-lld_inference is built from a base image containing an Intel CPU optimized image of Tensorflow. A (modern) Intel CPU is expected, AMD probably won't work. The usage below has not been updated, the --gpus flag is not required.

You need to specify input and output directories using the -v flag to docker run. Also set the output directory to be world writable

mkdir in out && chmod 777 out
docker run --rm -u $(id -u)                             \
    -v $PWD/in:/incoming -v $PWD/out:/outgoing          \
    fnndsc/pl-lld_inference lld_inference               \
    /incoming /outgoing

Development

Build the Docker container:

docker build -t local/pl-lld_inference .

Run unit tests:

docker run --rm local/pl-lld_inference nosetests

For in-container debugging, mount the source directories appropriately:

cd pl-lld_inference
docker run -it --rm --gpus all                                                  \
    -v $PWD/LLDcode:/opt/conda/lib/python3.6/site-packages/LLDcode              \
    -v $PWD/lld_inference:/opt/conda/lib/python3.6/site-packages/lld_inference  \
    -v $PWD/in:/incoming -v $PWD/out:/outgoing                                  \
    local/pl-lld_inference lld_inference                                        \
    /incoming /outgoing

Examples

# Assume you have some *mha leg images... copy them to the input directory.
# Obviously adjust below as you see fit!
cd ~/some/dir
mkdir in out && chmod 777 out
cp *mha in
docker run --rm -u $(id -u) --gpus all                                          \
    -v $PWD/in:/incoming -v $PWD/out:/outgoing                                  \
    fnndsc/pl-lld_inference lld_inference                                       \
    /incoming /outgoing

_-30-_

FNNDSC/pl-lld_inference