ESTRO 2025 - Abstract Book

S3712

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

4. Akcay M, Etiz D, Celik O, et al. Evaluation of acute hematological toxicity by machine learning in gynecologic cancers using postoperative radiotherapy[J]. Indian journal of cancer, 2022, 59(2): 178-186.

264

Digital Poster Z-Rad: the swiss pocket knife for radiomics Maksym Fritsak 1,2 , Marta Bogowicz 3,1 , Diem Vuong 1 , Florian Dietsche 1 , Anja Joye 1 , Matthias Guckenberger 1,2 , Stephanie Tanadini-Lang 1 , Hubert S. Gabryś 1 1 Department of Radiation Oncology, University Hospital Zurich and University of Zurich, Zurich, Switzerland. 2 Faculty of Medicine, University of Zurich, Zurich, Switzerland. 3 Department of Radiation Oncology (Maastro), GROW School for Oncology and Reproduction, Maastricht University Medical Centre+, Maastricht, Netherlands Purpose/Objective: Recent advancements in medical imaging have led to increased data availability and multi-center imaging biomarker studies, highlighting the need for standardized image analysis. Radiomics has the potential to advance precision medicine by improving diagnostic, prognostic, and predictive accuracy. However, radiomics applications are often limited by software requiring programming skills, by only partial compliance with the Image Biomarker Standardization Initiative (IBSI) [1,2], and by lack of comprehensive data-processing pipelines compatible with clinical data formats. To address these limitations, we developed Z-Rad [3], an open-source, user-friendly software designed to enable radiomics analysis for medical professionals without requiring programming expertise, while ensuring full IBSI compliance. Material/Methods: Z-Rad features a graphical user interface (GUI) and an application programming interface (API) in Python. It supports CT, PET, and MR data modalities in DICOM and NIfTI formats and offers comprehensive image preprocessing, image filtering, and radiomics extraction capabilities. Preprocessing provides 3D or 2D (axial slice-wise) resampling using nearest-neighbor, linear, B-spline, and Gaussian methods or conversion from DICOM to NIfTI format without resampling. Filtering includes mean filters, Laplacian of Gaussian, Laws kernels, and wavelets. Radiomics provides options for intensity range truncation, outlier exclusion, intensity discretization by fixed bin width or bin count, and six texture feature aggregation methods for 2D, 2.5D, and 3D. Results: Z-Rad demonstrates robust capabilities across multiple domains compared to other radiomics extraction tools, such as PyRadiomics, Medical Imaging Toolbox™, and SlicerRadiomics (Table 1). Unlike PyRadiomics and Medical Imaging Toolbox™, Z-Rad is accessible to users without programming skills. It provides modality-specific DICOM preprocessing for MRI, CT, and PET, a feature absent in PyRadiomics and only partially supported in SlicerRadiomics and Medical Imaging Toolbox™. Z-Rad is the only tool in this comparison that fully adheres to both IBSI I andI II, ensuring reliable and standardized feature extraction. Additionally, Z-Rad supports 2D, 2.5D, and 3D feature aggregation strategies, as well as built-in functionalities for intensity range re-segmentation and outlier filtering, features not present in PyRadiomics or SlicerRadiomics. Z-Rad has a significant advantage in PET image support compared to both commercial and open-source software (Table 2). Z-Rad incorporates a vendor-specific strategy that allows reliable conversion of the raw PET DICOM image into body weight-normalized SUV when MIM and 3D Slicer fail.