Detection and severity quantification of pulmonary embolism with 3D CT data using an automated deep learning-based...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Imaging

Background:

• Pulmonary embolism (PE) diagnosis relies on computed tomography pulmonary angiography (CTPA).
• Accurate quantification of PE severity is crucial for patient management.
• Current methods may require extensive manual annotation, increasing workload.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) model for detecting pulmonary embolism (PE).
• To quantify PE severity using the Qanadli score and right-to-left ventricle (RV/LV) diameter ratio.
• To utilize limited annotations on 3D CTPA scans for model training.

Main Methods:

• A DL pipeline was developed for blood clot detection, PE classification, Qanadli score estimation, and RV/LV ratio prediction.
• The model was trained on a large dataset of 3D CTPA examinations with image-level annotations.
• Performance was validated on independent datasets using Area Under the Curve (AUC) for classification and R² for quantification.

Main Results:

• The DL model achieved an AUC of 0.852 on the test set for PE classification.
• The model demonstrated strong performance in estimating PE severity, with R² values of 0.717 for the Qanadli score and 0.723 for the RV/LV ratio.
• These results indicate the model's capability in both detecting PE and quantifying its severity.

Conclusions:

• Deep learning-based AI tools can effectively detect pulmonary embolism and estimate its severity from 3D CTPA.
• The proposed method leverages automated blood clot and cardiac segmentation for analysis.
• Further clinical validation is recommended to assess the real-world impact of these AI tools.