Predicting Short-Term Mortality in Patients With Acute Pulmonary Embolism With Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Medicine

Background:

• Accurate short-term mortality prediction in acute pulmonary embolism (PE) is crucial for patient management.
• The Pulmonary Embolism Severity Index (PESI) is the current standard but has limitations in predictive accuracy.
• Developing advanced prediction models is essential for improving outcomes in PE patients.

Purpose of the Study:

• To develop and validate a novel deep learning (DL) model for predicting short-term mortality in acute PE.
• To utilize multimodal data, including imaging and clinical information, for enhanced predictive power.
• To compare the performance of the new DL model against the established PESI score.

Main Methods:

• A multimodal deep learning (mmDL) model was developed using contrast-enhanced multidetector computed tomography (MDCT) scans.
• Clinical and demographic data were integrated with imaging data for model training.
• Performance was evaluated using the area under the receiver operating characteristic curve (AUC) and compared to the PESI score.

Main Results:

• The mmDL model achieved a high AUC of 0.98, demonstrating excellent predictive accuracy.
• The PESI score achieved an AUC of 0.86.
• The mmDL model significantly outperformed the PESI score in predicting short-term mortality (P<0.001).

Conclusions:

• The developed mmDL model offers a highly accurate method for predicting short-term mortality in acute PE patients.
• This novel approach significantly surpasses the predictive capabilities of the current PESI score.
• The findings support the integration of multimodal deep learning in clinical practice for PE risk stratification.