MP45-17: Correlation between CTC counts and radiomic metrics from multiple osseous metastatic lesions in metastatic prostate cancer

Bino Varghese, Steven Cen, Gareth Morrison, Jonathan Buckley, Dejerianne Ostrow, David Quinn, Timothy Triche, Amir Goldkorn, Vinay Duddalwar*, Los Angeles, CA

Introduction: Identify CT radiomic correlates for liquid biopsy metrics such as circulating tumor cells (CTC) counts in patients with metastatic prostate cancer (PC). Together, CT radiomics and liquid biopsy could detect clonal heterogeneity and aid in improved patient management.

Methods: Under IRB-approved informed consent, blood was collected from 22 patients with metastatic castrate resistant PC (mCRPC). CTC counts were enumerated using CellSearch platforms. Five osseous metastatic lesions were manually segmented from CT scans in the same patients using ITK-SNAP by an experienced radiologist. Cancer imaging phenomics toolkit (CaPTk) was used for radiomics analysis. A panel of 6 metrics representing first-order statistical measures: Intensity and second-order statistical measures: Greylevel cooccurrence matrix (GLCM) were calculated from each segmented lesion per patient.  Mixed model correlation was used to examine correlation between weighted average of radiomic features from the five bone lesions per patient and their CTC counts.  Benjamini-Hochberg Procedure was used to control false discovery rate from multiple comparisons.

Results: The distribution of CTC counts was 102±198 (min: 0, max: 692). Among the 69 texture features extracted from the bone lesions, the weighted average of Intensity-based inter-quartile-range and GLCM-based kurtosis, variance and standard-deviation extracted from the 5 bone lesions per patient showed statistically significant correlation with CellSearch enumerated CTC counts ranged between 0.71 to 0.79 with adjusted p-value < 0.01). Radiomics and liquid biopsy have theoretically comparable advantages: they are both noninvasive, can be quantified and followed serially to evaluate disease progression. While their exact role in clinical practice is yet to be established, we conducted a preliminary integrative assessment of the two, within a small cohort. We show that the weighted average of select radiomic metrics extracted from multiple bone lesions (N=5) provide strong correlates for CTC counts.

Conclusions: Post validation in larger cohorts, radiomic corelates of liquid biopsy parameters such as CTC counts may provide surrogates for assessment of the likelihood of recurrence and metastasis in PC. With the widespread applicability of artificial intelligence in healthcare, diverse information about tumor behavior based on its genotype and phenotype decoded via imaging, may predict treatment failures earlier than with conventional methods.

Source of Funding: NCI: 1R01CA257610-01