ESTRO 2025 - Abstract Book

S3820

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

3373

Digital Poster 3D Printing a Textured Radiomics Phantom for CT Scanner Analysis Peter D McHale 1 , Owen McLaughlin 1 , Kathryn H Brown 1 , Karl T Butterworth 1 , Conor K McGarry 1,2 1 Patrick G. Johnston Centre for Cancer Research, Queen's University Belfast, Belfast, United Kingdom. 2 Northern Ireland Cancer Centre, Belfast Health & Social Care Trust, Belfast, United Kingdom Purpose/Objective: The purpose of this study was to replicate the credence cartridge radiomics (CCR) phantom [1] using 3D printing and verify its properties compared to the original phantom. Material/Methods: CT scans of the CCR were retrieved from the Cancer Imaging Archive (TCIA). Each cartridge was segmented using a 100.0x100.0x32.0 mm 3 volume to retrieve a mean CT number (HU) and standard deviation for phantom development. Markedly heterogeneous cartridges (rubber, wood) were recreated using multiple segments with varying infill or material (Table 1). Ultimaker S5 & S7 3D printers were used for prototyping; six filament materials were selected to approximate the original cartridges’ texture and HU. Infill densities (ID) was selected to reproduce the mean HU value for each original cartridge, except for acrylonite butadiene styrene (ABS), which was printed with the same ID specified for the original phantom (Table 1). Each cartridge was printed with outer dimensions of 40x40x10 mm 3 . A sleeve was printed (with polylactic acid) to contain the cartridges with outer dimensions of 100.0x100.0x100 mm 3 . The 3D printed and original CCR phantom were scanned on a Siemens SOMATOM Confidence (exposure: 160 mAs, voltage: 120 kV, slice thickness: 1 mm). Both were scanned twice for test-retest analysis using intraclass correlation coefficients (ICC) of radiomic features. An ICC above 0.8 was considered to show good feature reliability [2]. Four 2 cc segmentations were contoured in each cartridge and 107 radiomics features were extracted using PyRadiomics (v.8426f). Shape features (n=14) were removed from the analysis.