Room: Track 2
The variability in CT-based texture/radiomic characterization of tumors remains an unresolved issue. Methods to verify radiomic signature values are needed. We report on our efforts to design, fabricate, and test a semi-anthropomorphic 3D printed radiomic phantom to harmonize radiomic signatures.
CT scans of six patients, four with pancreatic ductal adenocarcinoma (PDAC), one with non-small cell lung carcinoma (NSCLC), and one with advanced stage hepatic cirrhosis were selected based on tissue heterogeneity. Segmented tumors were embedded into a heterogenous portion of the cirrhotic liver. The composite volume was then dithered, and 3D printed. Thirty repeat scans were acquired without movement between scans. Five additional scans were acquired using different scan modes. The similarity between the 3D print and tissue it was modeled after was assessed using the structural similarity index (SSIM). Prognostic radiomic signatures for NSCLC and PDAC were extracted. Statistical tests included the within-subject coefficient of variation (wCV), percent deviation (pDV), and one-sample Wilcoxon signed-rank test.
SSIM was 0.71. All prognostic NSCLC and PDACs radiomic signatures had wCV < 1.0%. The pDV of NSCLC energy was ± 1.0% (p = 0.290). NSCLC gray level non-uniformity from run-length matrices, pDV > -10% when edge enhancing kernels were used and pDV > 50% with increased pixel size. PDAC gray level co-occurrence matrix contrast and dissimilarity deviated by > ± 10% when 100 mA, DECT, 100 kVp, increased pixel size, and additional reconstruction methods were used. Across all scan modes, PDAC energy pDV ± 13%. At a reduced tube current, PDAC entropy pDV (%) > 15%.
We demonstrate a process to directly 3D print anatomy seen on CT scans while retaining the original structure and attenuation characteristics. The phantoms can then be used as the first step towards the clinical standardization and validation of quantitative radiomic signatures.
Funding Support, Disclosures, and Conflict of Interest: This research was funded in part through the NIH/NCI Cancer Center Support Grant P30 CA008748.