Skip to main content

Tool for automatically segmenting muscle and fat tissue in CT images acquired at the 3rd vertebral level

Project description

L3 Auto-segmentation Tool

Tool for automatically segmenting muscle and fat tissue in CT images acquired at the 3rd vertebral level

The AutoSegL3 tool allows a data manager to train a deep learning model that automatically segments muscle and fat tissue in CT images taken at the 3rd vertebral (L3) level. To train the deep learning model the tool needs a collection of L3 images and corresponding TAG files that contain the labels of each tissue to be segmented. To run the trained model on previously unseen CT images the tool only needs a collection of L3 images. The tool will then produce a mask for each L3 image that outlines the location of the muscle and fat regions.

For training, if default parameters are used, all the data manager has to do is point the tool to a directory containing L3 images and corresponding TAG files. From this directory, an HDF5 file will be generated. During this process the images and TAG files will be checked for certain characteristics like a common dimension of 512 by 512 pixels, pixels containing normalized Hounsfield units, etc. Any images that do pass this initial quality check will be reported in a text file.

For testing the training procedure, the tool also has to be pointed to a directory containing both L3 images and TAG files. However, only the L3 images will be used for generating segmentations. The TAG files will be used to evaluate the quality of the segmentations. This step will also produce a summary report containing some performance metrics, e.g., Dice scores. Note that the testing phase is only meant to obtain realistic performance metrics. To use the model for prediction, train it on all data you have (see next section).

For model preparation, train it on all data you have. Generate a CSV database containing certain clinical scores for each L3 image, e.g., SMRA, muscle index, SAT index and VAT index (what other scores can we think of?). This database can then be used to visualize the spread of scores across all images in the training data. When a new image is predicted you can also highlight its position within the spread of the training scores.

For prediction, the tool has to be pointed to a directory containing only L3 images.

Features

  • TODO

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

History

0.1.0 (2021-02-04)

  • First release on PyPI.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

autosegl3-0.11.0.tar.gz (19.9 kB view hashes)

Uploaded Source

Built Distribution

autosegl3-0.11.0-py2.py3-none-any.whl (13.9 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page