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

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Acute pulmonary embolism (PE) requires accurate assessment of lung perfusion defects (PDs).
• Dual-energy computed tomography (DECT) offers quantitative imaging capabilities for PE evaluation.
• Automated and semi-automated methods for PD quantification are under development.

Purpose of the Study:

• To evaluate the performance of automated DECT-based PD quantification in acute PE.
• To compare automated AI-based segmentation with semi-automatic clinician-guided segmentation.
• To examine the correlation of PD quantification with clinical parameters in acute PE patients.

Main Methods:

• Retrospective analysis of 171 patients with moderate-to-severe acute PE undergoing DECT.
• Quantification of PDs using a fully automated AI segmentation on iodine maps.
• Comparison with a semi-automatic method involving manual threshold adjustment by radiologists.
• Correlation with clinical variables: Miller score, RV/LV ratio, oxygen saturation, and symptoms.

Main Results:

• The semi-automatic method showed stronger correlations with embolic burden (Miller score, r=0.4) and oxygen saturation (r=-0.2).
• Fully automated AI quantification yielded lower PD values.
• Automated AI demonstrated weaker associations with clinical parameters compared to the semi-automatic method.

Conclusions:

• Semiautomatic quantification of lung PDs in acute PE offers superior accuracy and clinical relevance currently.
• Fully automated AI methods require further development for comparable performance.
• Future multimodal AI incorporating anatomical and clinical data may enhance diagnostic precision in PE.