Computer-aided detection system for pulmonary embolism with integrated cardiac ssessment based on embolic burden

Luque Del Toro, Iván

Computer-aided detection system for pulmonary embolism with integrated cardiac ssessment based on embolic burden

License

Citation

Luque Del Toro, I. Computer-aided detection system for pulmonary embolism with integrated cardiac ssessment based on embolic burden. 2023. handle: http://hdl.handle.net/10230/57960

This citation was generated automatically.

Abstract

Pulmonary embolism (PE) is a cardiovascular disease caused by one or several occlusions in the pulmonary arteries. Its diagnosis is mainly reliant on imaging, being computerized tomography pulmonary angiogram the gold standard. Recently, there has been increasing interest in automatizing PE detection with the use of computeraided detection systems, aiming to reduce workloads and enhance identification. Semiquantitative scores of embolic burden have also been proposed to characterize PE severity for better patient management. Yet, few attempts have been done to couple both. Here, we propose a system capable of PE detection, which exploits the visual explanations of the detector part to produce 2D representations of embolic burden. These are later used to predict right-to-left ventricle diameter (RV/LV) ratio ≥ 1, a prognosis cardiac feature strongly associated with embolic burden. The detector part is based on a Squeeze-and-Excitation-ResNet50, trained on a subset of the RSNA-STR Pulmonary Embolism CT dataset. The model achieves an accuracy of 0.72, sensitivity of 0.73, and specificity of 0.82 on the test set, which is slightly below the performance of radiologists. As the cardiac assessment module directly depends on the detector’s performance, we were unable to predict RV/LV ratio ≥ 1 successfully. Nevertheless, we believe our system is theoretically feasible and could assist in both PE detection and risk assessment in the future. For that, further work should focus on improving the performance of the detection model, especially regarding high false positive rates, and tune the assessment module accordingly.