Utilizing radiomic features of arterial phase computed tomography for delineating parathyroid adenomas from surrounding anatomical structures

Abstract

Aim: The study aimed to correlate radiomic data of four-dimensional computed tomography (4D-CT) with pathology-proven parathyroid adenomas to identify and quantitate select dimensional and textural features that predict parathyroid adenomas with a high degree of confidence, with the ultimate goal of improving the reliability of parathyroid adenoma detection so as to facilitate the subsequent unilateral minimally invasive parathyroidectomy (MIP). Methods: A total of 144 subjects with a history of neck 4D-CT, parathyroidectomy, and intraoperative pathology-proven parathyroid adenoma(s) were retrospectively reviewed. Following the exclusion of patients with a thyroidectomy, unsuccessful surgery, or indeterminate localization of the parathyroid adenoma on 4D-CT, a preliminary sample of 20 patients was obtained. Four anatomical structures (carotid artery, internal jugular vein, thyroid, and parathyroid adenoma) were segmented twice on 25-second arterial phase axial sections of a 4D-CT, and radiomic data of the shape, first-order, and second-order classes (106 variables) were extracted from the four structures for each patient. Results: Select radiomic variables among the carotid artery, jugular vein, and thyroid groups exhibited overall significant differences when compared to the parathyroid adenoma data (P < 0.05). Further Tukey’s post hoc analysis revealed that, when the parathyroid adenoma group was treated as the reference, 11/16 shape class, 16/18 first-order class, and 46/69 second-order class variables significantly differ from the carotid artery, jugular vein, and/or thyroid group(s). In addition, we found that the thyroid has distinct textural features compared to the parathyroid group, with 1/18 first-order and 19/69 second-order variables differing significantly between the two (P < 0.05). Notably, the texture variables such as dependence non-uniformity, long run emphasis, run percentage, run variance, and busyness exhibited the highest level of differences between the two groups (P < 0.0001). Conclusion: The parathyroid adenoma group is associated with a unique set of radiomic variables in comparison to surrounding anatomy such as the carotid artery, internal jugular vein, and thyroid. Relevance for Patients: The distinct, quantifiable differences in dimensional and textural features serve as a set of signature markers distinguishing parathyroid adenomas from their surrounding structures in 4D-CT. These attributes obviate the need for invasively locating parathyroid adenomas preoperatively, thereby enhancing the utilization rate of MIP, which has a favorable implication in the overall clinical outcomes.