Abstract 1070: Predicting Short Term Mortality In Patients With Acute Pulmonary Embolism With Deep Learning

Background: Pulmonary embolism (PE) is one of the leading causes of short-term mortality among cardiovascular diseases. Rapid and accurate stratification of PE-related risk is vital for effective patient management. A strong potential research in this direction is to utilize deep learning (DL) approaches. DL approaches have enhanced risk prediction performance by leveraging novel risk predictors and analyzing complex interrelationships among them. In this study, we developed a DL based new scoring system superior to the current standard, the Pulmonary Embolism Severity Index (PESI) , when predicting the short-term (0-30 days) mortality of PE. Materials and Methods: Our study included a total of 207 patients. The diagnosis of acute PE was confirmed using contrast-enhanced computed tomography (CT) scans. DL model was created using CT scans and corresponding ground truth labels (short term mortality). We have benchmarked different DL architectures for this purpose including CNN and Transformers. Results: Out of 207 patients, 53 of them died during short term . The research group was categorized into two groups based on their short term survival status. Multivariable logistic regression analysis was used to determine the independent relationships between variables and short term mortality. Short-term mortality was predicted statistically significantly (p<0.001) in the DL model developed from the patient's CT images (Figure 1). Then, we developed a new scoring model by uniquely combining the short-term mortality independent predictors, baseline data and the DL model (Figure 1). The ROC analysis demonstrated an AUC of 0.976 (p<0.001) for the Baseline-DL model and an AUC of 0.865 (p<0.001) for the PESI score The Baseline-DL model was statistically significantly superior to the PESI (p<0.001). Conclusion: We developed a DL based new scoring system to predict short term mortality of patients with PE, significantly superior to the PESI.