Role of radiomics to differentiate benign from malignant pheochromocytomas and paragangliomas on contrast enhanced CT scans.

Abstract

e14596 Background: Radiomics features, which are quantitative features generated by computational analysis of routine clinical imaging like CT scans, have been shown to be associated with clinical outcomes and tumor’s behavior in some solid tumors. We compared the radiomic features of malignant and benign pheochromocytomas/paragangliomas (P/P). Methods: Through an IRB approved study at our institution, we identified 20 consecutive patients with P/P and with available contrast-enhanced abdominopelvic CT. A radiologist with experience in oncologic imaging identified and segmented tumors on every slice using a MatLab-based imaging platform. The entire tumor image then underwent computational analysis generating 1160 radiomics features reflecting tumor size, shape, density, textural heterogeneity, and margins. These radiomics features were compared between malignant and benign P/P using Wilcoxon-Rank sum test. Results: Of the twenty patients included in this analysis, there were 6 patients with malignant P/P and 14 patients with benign tumors. Patients had been followed for at least 5 and many for at least 10 years after resection of the tumor. At diagnosis, the mean age of patients with benign and malignant tumors were 51 and 45, respectively. A 60% majority of patients with benign tumors were females while a 77% majority of patients with malignant tumors were male. Benign P/P were significantly different from malignant ones in: tumor intensity textures (spatial correlation [p-value = 0.0010], Laws [p-value = 0.0064], LoG [p-value = 0.0087], and Gabor [p-value = 0.0325]), and tumor local surface shape (Shape Index SI7 [p-value = 0.0325]). Conclusions: This initial analysis sought to discern differences in these rare tumors that might be exploited clinically. The results show that compared to benign tumors, malignant P/P tend to have more heterogenous texture, irregular edges, and less rounded shape on contrast enhanced abdominal CT scan. However, because these radiomics phenotype properties are subtle, they cannot be made reliably in an objective fashion using human visual assessment and thus these radiomics features may have a role as a quantitative imaging biomarker in P/P to predict tumor behavior. The cohort is being expanded and data will be updated at the time of the presentation. With larger numbers, the contribution to the radiomics profile of a SDHx mutation will be explored in greater depth to understand the differential impact of SDHx loss and of evolution into a cancer to the radiomics profiles.