MLX-Radiomics: accelerating standardized regional and voxel-wise radiomics for CT, PET and MRI

MLX-Radiomics agreed with all 2,160 IBSI reference measurements within specified tolerances and produced voxel-wise feature maps up to 218× faster than PyRadiomics on CT crops.

• Why it matters: Radiomics implementations vary, making cross-study comparisons difficult, and voxel-wise feature extraction is computationally demanding. Standardized, fast regional and voxel-wise mapping is needed for quantitative CT, PET, and MRI research.
• What they did: The tool implements 172 IBSI feature definitions plus 16 extensions and IBSI 2 image filters, then tested numerical accuracy against IBSI reference data and independent double-precision/analytic calculations. It benchmarked extraction time and memory on Apple M1 Max and M5 Max versus PyRadiomics, MIRP, Flash-Radiomics, and FastRad, and compared voxel-wise map generation for regions up to 10.1 million voxels.
• The result: In matched comparisons, it was 2.66–7.27× faster than PyRadiomics and MIRP for regional features and 198–218× faster than PyRadiomics for voxel-wise mapping on a 24×24×24 crop. In a 51-patient CT, FDG PET, and T1-weighted MRI cohort, it agreed with MIRP on 98.0% of requested values, with edge-tumor agreement improved by matching resampling rules.

AI summary of the abstract · may contain errors · not medical advice · read the paper.