Code and results: radiomic texture features under lossy compression

Abstract

Code and result files accompanying the manuscript “Radiomic Texture Features Can Survive Lossy Compression Exactly: Fixed-Bin-Size Discretisation Is a Threshold Panel” (I. Avcibas, 2026), prepared for submission to the Journal of Imaging Informatics in Medicine. Radiomic texture features are computed from discretised grey levels, and under a fixed bin size with a fixed origin a voxel’s grey level is determined by comparisons of its value with the bin edges. A lossy representation whose quantisation cells straddle no bin edge therefore reproduces every such feature exactly, at any error. This deposit measures that with pyradiomics on 200 tumour and liver regions of two tasks of the Medical Segmentation Decathlon under nine lossy representations, measures the bit rate of the aligned grids, and settles what happens when the pipeline interpolates to isotropic voxels before discretising. It holds the analysis scripts, the statements fixed in advance in their docstrings, and every result file and run log behind the tables and figures, including the statement that failed. The CT images are public (medicaldecathlon.com) and are not redistributed here.