Assessment of threshold-based infarct core and penumbra segmentations from spectral CT angiography iodine density maps

Abstract

Abstract CT perfusion (CTP) can differentiate infarct core (IC), penumbra (PEN), and healthy area (HA) in acute ischemic stroke. Quantitative iodine density maps from spectral CT angiography (CTA) may offer a lower-dose alternative. This study evaluated whether thresholding relative iodine density (rID) can reproduce CTP-derived IC and PEN segmentations. Thirty-nine patients with proximal middle cerebral artery occlusion undergoing spectral non-contrast CT, CTP, and CTA were included. rID thresholds differentiating IC, PEN, and HA were determined by region-level Youden-index and voxel-level Dice score optimization. rID-thresholded volumes were compared with CTP-derived volumes using Bland–Altman analysis and Pearson correlation. Region-level optimization yielded significant separation between tissue classes with rID thresholds of 0.58 (IC/PEN) and 0.83 (PEN/HA). However, voxel-level thresholding overestimated IC volumes (bias = 110 ml) and underestimated PEN volumes (bias = − 39 ml). Dice-optimized thresholds of 0.32 (IC/PEN) and 0.67 (PEN/HA) markedly reduced IC bias (− 4 ml) and showed moderate correlation with CTP-derived IC volumes ( r = 0.63), although limits of agreement remained wide (± 132 ml). PEN volumes showed no correlation with CTP-derived volumes. These exploratory findings suggest that spectral CTA has potential for IC estimation, but the current wide limits of agreement rule out reliable individual-patient use, warranting methodological improvement.