Abstract Rationale Accurate pulmonary embolism (PE) diagnosis traditionally relies on a combination of clinical assessment and imaging modalities, most commonly CT pulmonary angiography (CTPA) or ventilation-perfusion (V/Q) scanning. A recently developed technique, ventilation and pulmonary perfusion imaging from non-contrast CT (CT:VQ, 4DMedical™), offers a radio tracer- and contrast-sparing alternative for evaluating regional lung perfusion. This study investigates the capability of CT:VQ-derived perfusion images to automatically identify subjects with PE. Methods In this preliminary study, subjects with radiologist-confirmed PE (n = 12), control subjects (n = 17), and subjects with chronic obstructive pulmonary disease (COPD) (n = 29) were included. All participants provided written informed consent. Each subject underwent paired inspiratory and expiratory breath-hold CT scans, which were used to generate pulmonary perfusion images via CT:VQ. Perfusion defects were automatically identified from the CT:VQ images, and a set of defect-level features, including shape, anatomical location, contrast, and ventilation-perfusion matching, was extracted. A logistic regression model was trained to estimate the probability of PE for each defect. These defect-level probabilities were then aggregated to classify each subject as either PE or non-PE. Results The PE detection model achieved an area under the receiver operating characteristic curve (AUC) of 0.90 and correctly identified all 12 subjects with PE, corresponding to 100% sensitivity. Among the control and COPD groups, 32 subjects were correctly classified as non-PE, yielding a specificity of 68%. Three-dimensional visualizations highlighting the predicted PE defect locations were generated for each subject. Conclusions Automated detection of pulmonary embolism using CT:VQ perfusion images demonstrated high sensitivity with promising diagnostic performance. This contrast-free, radio tracer-free imaging method offers several advantages over current standards of care, including rapid processing and the ability to localize individual embolic defects within localized lung regions. Figure 1. Example from a subject with a confirmed right middle lobe PE (red arrows point to the PE). A Nuclear medicine SPECT coronal slice displaying the PE defect. B CT:VQ coronal slice displaying the PE defect. C 3D perfusion render from CT:VQ. D 3D render displaying the location (red) of the detected PE within the lung. This abstract is funded by: 4D Medical
Nilsen et al. (Fri,) studied this question.