Differentiation of low-attenuation intracranial hemorrhage and calcification using dual-energy computed tomography in a phantom system

Objectives: Calcific and hemorrhagic intracranial lesions with attenuation levels of less than 100 Hounsfield units (HUs) cannot currently be reliably differentiated by single-energy computed tomography (SECT). The proper differentiation of these lesion types would have a multitude of clinical applications. A phantommodelwas used to test the ability of dual-energy CT (DECT) to differentiate such lesions. Materials and Methods: Agar gel-bound ferric oxide and hydroxyapatite were used to model hemorrhage and calcification, respectively. Gel models were scanned using SECT and DECT and organized into SECT attenuationmatched pairs at 16 attenuation levels between 0 and 100 HU. Dual-energy CT data were analyzed using 3-dimensional (3D) Gaussian mixture models (GMMs), as well as a simplified threshold plane metric derived from the 3D GMM, to assign voxels to hemorrhagic or calcific categories. Accuracy was calculated by comparing predicted voxel assignmentswith actual voxel identities. Results: We measured 6032 voxels from each gel model, for a total of 193,024 data points (16 matched model pairs). Both the 3D GMM and its more clinically implementable threshold plane derivative yielded similar results, with higher than 90% accuracy at matched SECT attenuation levels of 50 HU and greater. Conclusions: Hemorrhagic and calcific lesions with attenuation levels between 50 and 100 HU were differentiable using DECT in a clinically relevant phantom system with higher than 90% accuracy. This method warrants further testing for potential clinical applications.