Improved detection of small pulmonary embolism on unenhanced computed tomography using an artificial...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Pulmonary embolism (PE) diagnosis typically relies on CT pulmonary angiography (CTA).
• Unenhanced chest CT scans are common in oncological patients, offering an opportunity for incidental PE detection.
• Accurate localization of PE is crucial for treatment decisions.

Purpose of the Study:

• To evaluate the feasibility of a deep learning (DL) artificial intelligence (AI) model for localizing pulmonary embolism (PE) on unenhanced chest CT.
• To compare the AI model's performance against the gold standard of contrast-enhanced CT pulmonary angiography (CTA).

Main Methods:

• Retrospective analysis of 99 oncological patients with incidentally diagnosed PE on both unenhanced and contrast-enhanced chest CT.
• Training and testing a DL AI algorithm on unenhanced chest CT datasets.
• Post-processing AI model outputs by assessing intersection with lung segmentation.

Main Results:

• The AI algorithm achieved an overall sensitivity of 54.5% for central, 81.9% for segmental, and 80.0% for subsegmental PE (with 20 candidate boxes).
• Detection rates with a single box were 18.1% (central), 34.7% (segmental), and 0.0% (subsegmental).
• Median clot volumes differed significantly across central, segmental, and subsegmental PE locations (p < 0.05).

Conclusions:

• The developed AI algorithm demonstrates high sensitivity for detecting PE, particularly in segmental and subsegmental locations.
• This AI tool may assist in determining the necessity of contrast-enhanced CT for PE diagnosis.
• Further validation is warranted to establish the clinical utility of AI in PE detection on unenhanced CT.