ARRS 2022 Abstracts

RETURN TO ABSTRACT LISTING


1263. Assessment of Inter-Lesion Heterogeneity in Radiomic-Profiles of Metastatic Sites in Prostate and Ovarian Cancers
Authors * Denotes Presenting Author
  1. Bino Varghese; Keck School of Medicine at USC
  2. Redmond-Craig Anderson *; Keck School of Medicine at USC
  3. Steven Cen; Keck School of Medicine at USC
  4. Xiaomeng Lei; Keck School of Medicine at USC
  5. Darryl Hwang; Keck School of Medicine at USC
  6. Patrick Colletti; Keck School of Medicine at USC
  7. Vinay Duddalwar; Keck School of Medicine at USC
Objective:
Tumor heterogeneity has important clinical implications including treatment and prognostication. The demonstration of significant heterogeneity in radiomic profiles of different metastatic lesions in the same organ of a patient indicate the need for multiple regions of interest (ROIs) to prevent site selection biases in radiomic analysis. Using CT-based textural metrics (CTTA), we evaluate the inter-lesion heterogeneity in radiomic profiles that exist between metastatic lesions in bone found in prostate cancer (PCa) and peritoneal metastases found in ovarian cancer (OV).

Materials and Methods:
In this IRB-approved, HIPAA-compliant, retrospective study, we evaluated CT studies from 22 patients with PCa and 14 patients with OV. For all patients, at least five bone metastatic lesions in PCa and at least five peritoneal metastases in OV were digitally segmented by an experienced radiologist on Synapse3D software. The Cancer Imaging Phenomics (CAPtk) Toolkit was used for radiomic analysis. A total of 1360 texture metrics were extracted from each metastatic lesion. A mixed-effect model was used for variance component analysis. We estimated the inter-patient and intra-patient variance (random effect), then calculated the percentage of intra-patient variance (IPV) that contributed to the total variance. A Z test was used to assess the random effect was 0. Histograms were used to display the percent variance components across radiomic features.

Results:
The radiomic profile is more heterogenous across the lesions from the same patient in PCa, and more homogenous across lesions from the same patient in OV. For OV patients, there were only 4.68% features with percent IPV>10%. Only 0.58% of total features were noted with IPV statistically significantly larger than 0. In contrast, prostate cancer has 48.5% features with percent IPV>10%, and 51.5% of total features with IPV statistically significantly larger than 0. The mean, standard deviation, interquartile range of percent IPV for OV and PCa were 17.3%±22.3 (0.45%-21.3%), 3.2%±2.1 (0.46%-0.56%), respectively.

Conclusion:
In radiomic evaluation of patients with multiple lesions in the same organ, multiple metastatic lesion ROIs may need to be evaluated to negate sampling site biases. This would vary depending on the primary tumor as well as the metastatic site(s). This data may assist with the need for ROI segmentation in future radiomic studies.