/deep-learning-models-for-the-differentiation-of-osteoporotic-and-malignant-vertebral-fractures

Primary LanguagePythonMIT LicenseMIT

Development and validation of CT-based deep learning models for the differentiation of osteoporotic and malignant vertebral fractures

This repository contains the code for distinguishing between malignat and osteoporotic fractures in CT scans in the nifti format.

Input Data Format

  • The dataset has to have three files corresponding to one data sample: image, segmentation mask, centroid annotations. The format of those files follows VerSe dataset format (https://github.com/anjany/verse)

  • Sub-directory-based arrangement for each patient. File names are constructed of entities, a suffix and a file extension following the conventions of the Brain Imaging Data Structure (BIDS; https://bids.neuroimaging.io/)

Example:
-------
sample_dataset/rawdata/sub-sample000
    sub-sample000_ct.nii.gz - CT image series

sample_dataset/derivatives/sub-sample000/
    sub-sample000_seg-vert_msk.nii.gz - Segmentation mask of the vertebrae
    sub-sample000_seg-subreg_ctd.json - Centroid coordinates in image space

Usage

python3 script.py  -path /path/to/image  -vert vert1 vert2 vert3
  • Sample Example
python3 script.py  -path ./sample_dataset/rawdata/sub-sample001/sub-sample001_ct.nii.gz  -vert L2

Requirements

  • Python 3.11.3
  • requirements.txt

Output format

  • An output json file is saved in the derivatives folder of the scan, containing the prediction (0-osteoportic, 1-malignant) and the uncertainty(0-certain, 1-uncertain)

References:

  1. .

Under Construction

  • Push inference code
  • [] Batch input support
  • [] Push training code