Artificial Intelligence-Assisted Lung Perfusion Quantification from Spectral CT Iodine Map in Pulmonary Embolism

<b>Introduction</b>: This study evaluated the performance of automated dual-energy computed tomography (DECT)-based quantification of perfusion defects (PDs) in acute pulmonary embolism and examined its correlation with clinical parameters. <b>Methods</b>: We retrospectively analyzed data from 171 patients treated for moderate-to-severe acute pulmonary embolism, who underwent DECT imaging at two separate time points. PDs were quantified using a fully automated AI-based segmentation method that relied exclusively on iodine perfusion maps. This was compared with a semi-automatic clinician-guided segmentation, where radiologists manually adjusted thresholds to eliminate artifacts. Clinical variables including the Miller obstruction score, right-to-left ventricular diameter ratio, oxygen saturation, and patient-reported symptoms were also collected. <b>Results</b>: The semiautomatic method demonstrated stronger correlations with embolic burden (Miller score; r = 0.4, <i>p</i> < 0.001 at follow-up) and a negative correlation with oxygen saturation (r = -0.2, <i>p</i> = 0.04). In contrast, the fully automated AI-based quantification consistently produced lower PD values and demonstrated weaker associations with clinical parameters. <b>Conclusions</b>: Semiautomatic quantification of PDs currently provides superior accuracy and clinical relevance for evaluating lung PDs in acute pulmonary embolism. Future multimodal AI models that incorporate both anatomical and clinical data may further enhance diagnostic precision.