Iodine Versus Calcium Quantification Via Analyzing HU Spectral Curves In Dual-Energy CT for Radiotherapy Planning
Abstract
Purpose
Iodine maps derived from Dual-Energy CT (DECT) provide critical biological information for radiotherapy treatment planning; however, conventional clinical iodine maps often mistakenly include bones due to insufficient X-ray spectral separation. In this study, we propose a robust method for iodine and calcium quantification using HU spectral curves from virtual mono-energetic images (VMIs), aiming to improve target delineation by accurately differentiating iodine-enhanced tumors from adjacent calcium-containing bone.
Methods
The proposed method used a multi-energy CT QA phantom to construct a voxel-clustering template for varying iodine and calcium concentrations. Principal component analysis (PCA) was applied to the HU spectral curves of all voxels in the phantom VMIs. Material groups were identified in the PCA space, where an iodine and calcium clustering template was fitted to characterize their respective concentrations. For applications, patient VMIs were projected into the same PCA space, the material concentrations were determined using the predefined clustering template. This approach was preliminarily evaluated in head and neck (HN) cancer patients undergoing iodine-enhanced DECT for radiotherapy simulation.
Results
The proposed approach successfully differentiated iodine and calcium voxels using a learned clustering template in PCA space. Unlike clinically available software, the calcium areas were successfully excluded from iodine maps, effectively minimizing false iodine signals from bone. When applied to clinical datasets of the HN patients, the method significantly improved the quantification and differentiation of iodine-enhanced tumor regions from adjacent bone structures.
Conclusion
The proposed approach enhances the visibility and differentiation of iodine-enhanced regions from bone structures, potentially leading to more robust target and organ-at-risk (OARs) contouring in radiotherapy planning. These results demonstrate the potential of using learned clustering templates for reliable material quantification and separation, supporting more precise anatomical delineation under complex clinical imaging conditions.