Pulmonary Embolism Survival Prediction Using Multimodal Learning Based on Computed Tomography Angiography and...

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Pulmonary embolism (PE) is a leading cause of mortality.
• Accurate survival prediction is crucial for managing PE patients.
• Current prediction methods may benefit from advanced analytical tools.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for predicting PE survival.
• To integrate computed tomography pulmonary angiography (CTPA) imaging features, clinical data, and PE Severity Index (PESI) scores.
• To compare the performance of multimodal DL models against traditional methods.

Main Methods:

• Retrospective analysis of 918 patients with 3978 CTPAs.
• AI model used to extract imaging features from CTPAs.
• Development of multimodal CoxPH models combining DL features, clinical data, and PESI scores.
• Performance evaluated using concordance index (c-index) and Kaplan-Meier analysis.

Main Results:

• Multimodal DL models incorporating CTPA features, clinical data, and PESI achieved higher c-indices than PESI alone.
• Stratification by models significantly differentiated survival outcomes (P<0.001).
• A strong correlation was observed between high-risk grouping and right ventricular (RV) dysfunction.

Conclusions:

• Multiomic DL models integrating CTPA, clinical data, and PESI offer superior PE survival prediction.
• These advanced models enhance risk stratification and can inform clinical decision-making.
• DL models show promise in improving prognostic accuracy for PE.