Pulmonary Embolism Detection in Unenhanced CT Exams: A Radiomic-Based Classifier as a Promising Screening-Diagnostic Tool.

Background

Pulmonary embolism (PE) is a life-threatening condition in which early diagnosis and prompt treatment are essential to reduce morbidity and mortality. Recent advances in radiomics have introduced software tools for PE detection on CT pulmonary angiography (CTPA), but their application to non-contrast chest CT remains unexplored.

Methods

We retrospectively identified 57 CTPA examinations performed between January 2020 and March 2023. Patients were randomly assigned to training and validation cohorts; an additional 14 PE-negative patients were included in the validation cohort. Spherical regions of interest (ROIs) were manually placed within emboli and within normal-flow pulmonary arteries. From each ROI, 836 radiomic features were extracted. Statistically significant features were identified using the Wilcoxon test. A logistic regression classifier was then developed to discriminate between embolus and control ROIs based on their radiomic signatures.

Results

Of the 836 features, 242 (all derived from wavelet-filtered images) showed significant differences between embolus and normal-flow ROIs on unenhanced CT. The classifier demonstrated high diagnostic performance in the validation cohort, with an AUC of 0.87 (95% CI: 0.86-0.88) and an accuracy of 0.77 (95% CI: 0.76-0.79), with only a limited number of misclassified ROIs.

Conclusions

This preliminary study demonstrates the potential of radiomics to identify thrombi on unenhanced chest CT scans, offering a promising avenue for the early detection of acute PE. The proposed model showed satisfactory diagnostic accuracy with low few misclassifications, supporting further validation in larger cohorts.