PET-LAB

This repository contains code for radiomics reproducibility analysis of the data from the TROG 12.02 PET-LABRADOR trial. An outline of the files is provided below:

CV-scatter

Generating stripplots (a type of scatter plot) for the coefficient of variation (CV) for radiomics features of different MRI sequences. The CV is a measure of dispersion of the data.

CV-table

Generates a dataframe of the median CV from dataframes of CV for all patients and timepoints studied (i.e. gets the median across timepoints).

ICC-heatmap

Plots heatmaps of intra-class correlation coefficients for radiomics features, for different radiomics classes (e.g. first order, shape-based, etc).

ME_image_generation

Code for generating Maximum Enhancement (ME) images from DCE image series.

Radiomics

Runs radiomics analysis on Before, During, and After-PST (Primary Systemic Therapy) DCE-MRI (ME), T1w-MRI (DCE pre-contrast), T2w-MRI, and ADC (DWI-MRI) images, including isotropic voxel resampling, z-score normalisation (only for T1w and T2w images), grey-level discretisation, and cropping to speed up processing. Note that the z-score normalisation does not work well for T2w images (unresolved issue).

Stats-across-timepoints

Calculate statistics on radiomics features across timepoints. Statistics include CV, wCV (within-subject CV from QIBA paper), relative difference from baseline and assessment of normality of relative differences using Shapiro-Wilk test, mean and standard deviation from the relative differences, and also calculate the intra-class correlation coefficient (ICC). QIBA paper: van Houdt PJ, Saeed H, Thorwarth D, Fuller CD, Hall WA, McDonald BA, Shukla-Dave A, Kooreman ES, Philippens ME, van Lier AL, Keesman R. Integration of quantitative imaging biomarkers in clinical trials for MR-guided radiotherapy: Conceptual guidance for multicentre studies from the MR-Linac Consortium Imaging Biomarker Working Group. European Journal of Cancer. 2021 Aug 1;153:64-71.

Stats-within-effects

Calculate statistics comparing between the different bin counts/normalisations rather than different time points (choose 1 time point) (CV and ICC). It probably makes more sense to look for changes across timepoints, as these will reflect biological/physical changes.

T2w_breast_mask_registration

Code to register T2w images (timepoint 1) to timepoint 2 and 3, and then after registering the images, code to propagate the transforms (on the masks). After the transforms are propagated, split the masks into left and right (L and R) breasts.

Tumour_delta

Obtain delta radiomics for radiomics of tumours.

Convert_and_process_data.ipynb

Code to process DICOMM contours to NIFTI (written by Rob Finnegan).

erode_masks.py

Code to erode the breast masks used for whole-breast radiomics (to remove suspected skin and muscle tissue from the analysis).

prepare_dicom.ipynb

Code to prepare dicom files for conversion to nifti by convert_and_process_data script. Code incomplete.

test_df

Test file for concatenating dataframes in pandas.

testing

Testing resampling code for different MRI sequences.

wes_014_tumour_code

Code to mask out second tumour for WES_014 patient tumour contour data.

DCE MRI Tumour segmentation testing.py

Code to test an alternative method of tumour segmentation based on the DCE sequence. This method did not appear to work for the Petlab data, as the tumour didn’t reach maximum intensity before the end of the scan time, a necessary feature of the method (unfortunately I have forgotten which paper’s method I was using).

Hist seg WES_006_4.py

Example code used in Honours project to generate tumour segmentations based on intensity thresholding

MRI breast registration.py

Code which can be used to register T2W images to B50T or DCE images using mutual information, then to propagate the transforms on breast masks and split them into L and R breast masks.

MRI registration.py

Code which can be used to register MR images within sequence types (e.g. DWI: b50 to b800), and T2w images to any other MRI sequence. After registering the image, this script also includes code to propagate the transforms on the masks and split them into L and R breasts.

Register-b50-to-b50.py

Code to register a b50 image from one patient to a b50 image of another patient (for standardisation of the data set)

Sphere generation WES_003_2.py

Code for an example patient (WES_003) and time point (2, during-PST), which generates a uniform sphere in the centre of the patient’s contralateral-to-tumour (healthy) breast. This can be used to compare the tumour with a similarly-sized region of healthy tissue from the same patient.

Test.py

Code to test image registration from T2w to DCE and DWI MRI sequences for a given patient and time point