Deep Learning-Based Algorithm for Automatic Detection of Pulmonary Embolism in Chest CT Angiograms

Abstract

Purpose: Since the prompt recognition of acute pulmonary embolism (PE) and the immediate initiation of treatment can significantly reduce the risk of death, we developed a deep learning (DL)-based application aimed to automatically detect PEs on chest computed tomography angiograms (CTAs) and alert radiologists for an urgent interpretation. Convolutional neural networks (CNNs) were used to design the application. The associated algorithm used a hybrid 3D/2D UNet topology. The training phase was performed on datasets adequately distributed in terms of vendors, patient age, slice thickness, and kVp. The objective of this study was to validate the performance of the algorithm in detecting suspected PEs on CTAs. Methods: The validation dataset included 387 anonymized real-world chest CTAs from multiple clinical sites (228 U.S. cities). The data were acquired on 41 different scanner models from five different scanner makers. The ground truth (presence or absence of PE on CTA images) was established by three independent U.S. board-certified radiologists. Results: The algorithm correctly identified 170 of 186 exams positive for PE (sensitivity 91.4% [95% CI: 86.4–95.0%]) and 184 of 201 exams negative for PE (specificity 91.5% [95% CI: 86.8–95.0%]), leading to an accuracy of 91.5%. False negative cases were either chronic PEs or PEs at the limit of subsegmental arteries and close to partial volume effect artifacts. Most of the false positive findings were due to contrast agent-related fluid artifacts, pulmonary veins, and lymph nodes. Conclusions: The DL-based algorithm has a high degree of diagnostic accuracy with balanced sensitivity and specificity for the detection of PE on CTAs.

New Diagnostic Tools for Pulmonary Embolism Detection

“…Prompt recognition of acute PE and immediate initiation of anticoagulation and mechanical thrombectomy can reduce the risk of death. 4,5 PE presentation can mimic a range of other conditions, and the most common cause of death from PE is failure to diagnose. While CTA is the imaging modality of choice to diagnose PE, its diagnosis is time-consuming and requires radiology expertise.…”

“…While CTA is the imaging modality of choice to diagnose PE, its diagnosis is time-consuming and requires radiology expertise. 4 Over the last two decades, the use of CTA has increased over 27-fold, increasing the workload of providers and radiologists and further complicating the PE workflow. AI trained for PE detection and RV/LV measurement paired with real-time collaboration tools is well-suited to address these issues.…”

“…AI trained for PE detection and RV/LV measurement paired with real-time collaboration tools is well-suited to address these issues. 4…”

“…Imaging-focused AI algorithms designed to evaluate and detect PE on CTA have a high level of accuracy. 4 In a matter of seconds, an AI algorithm evaluates a CTPA to quickly identify PE, including the size and location of an embolus. 7 Guidelines acknowledge that an RV/LV ratio > 0.9 is a marker of CTPA-based RV dysfunction and right heart strain caused by the PE, but the accuracy of reported measurements is inconsistent.…”

Cardiovascular Computed Tomography in the Diagnosis of Cardiovascular Disease: Beyond Lumen Assessment

“…Studies have reported promising results regarding the application of DL models for the automatic detection of PEs based on CTPA images [209][210][211][212]. A recent systematic review and meta-analysis of using DL in the detection of PEs showed a pooled sensitivity of 88% and a specificity of 86% based on an analysis of five studies [209].…”