Imaging

Session: Educational Exhibit Presentations - Station 8: Neurointerventional Radiology and Nonvascular Interventions

Detection of Right Heart Strain in Pulmonary Embolism Using Artificial Intelligence

Monday, March 6, 2023

5:45 PM – 5:50 PM

Location: Phoenix Convention Center, North Ballroom 120ABC, Poster Hall

Abstract Presenter(s)

AA

Alexander Anavim, MD

Integrated IR/DR Resident Physician
Einstein Medical Center, Jefferson Health

Disclosure(s): No financial relationships to disclose

Author/Co-author(s)

Akash Desai, MD

Integrated Resident
Einstein Health Care Network
Isael Perez, Jr., MD

Independent IR Resident Physician
Einstein Medical Center - Thomas Jefferson
Jung H. Yun, MD

Integrated IR/DR Resident Physician
Einstein Medical Center, Jefferson Health
SD

Sagar Desai, DO

IR/DR Resident
Einstein Medical Center
BN

Bala Natarajan, MD

Attending Interventional Radiologist
Einstein Medical Center, Jefferson Health
Paul Brady, MD (he/him/his)

Attending Physician, Director of Interventional Radiology
Albert Einstein Health Care Network

Disclosure(s):

Alexander Anavim, MD: No financial relationships to disclose

Akash Desai, MD: No financial relationships to disclose

Jung H. Yun, MD: No financial relationships to disclose

Paul Brady, MD: No financial relationships to disclose

Learning Objectives: To facilitate detection of right heart strain with pulmonary embolism (PE) using machine learning to expedite intervention.

Background:

Pulmonary embolism (PE) is a medical emergency representing the third most common cause of death worldwide, responsible for 100,000 deaths annually within the United States alone. PE has high mortality rates within the first few hours of presentation, thus, accurate and timely diagnosis is essential for initiating potentially lifesaving interventions. Management of PE has dramatically improved with recent advances in device technology, allowing improved percutaneous thrombectomy early on in management of select submassive and massive pulmonary emboli.

Artificial Intelligence (AI) powered algorithms are becoming more integrated within diagnostic radiology. The use of AI across multiple modalities and specialties within radiology is gradually changing clinical practice by enhancing accuracy and efficiency, interrater reliability, and overall workflow for more timely recommendations. AI algorithms have been shows to have a high sensitivity in detection of PE on CT pulmonary angiograms. Algorithms have also been developed to estimate right heart strain associated with PE. The purpose of our study was to determine accuracy of AI in detecting right heart strain in patients with PE.

Clinical Findings/Procedure Details:

CT pulmonary angiograms (CTPA) of patients presenting to a tertiary center emergency department between July 25, 2021 and October 16, 2021, were reviewed. A total of 74 patients with acute PE shown in CTPA were included. An AI algorithm was used to calculate right to left ventricular ratio. The calculated results of right ventricular strain by AI algorithm, the radiology reports, and echocardiogram results were collected. A right ventricle to left ventricle ratio of 1.2 was set as a trigger for the AI algorithm to detect right heart strain.

The result of right ventricular strain calculated by AI algorithm and radiology reports was concordant in 64 cases (20 positive, 44 negative cases) and discordant in 10 cases (6 positive and 4 negatives cases by AI). More than half of the patients (38/74 patients) had subsequent echocardiograms. Compared to the results of echocardiogram, the AI algorithm had a higher sensitivity (80% vs 73%) than the radiology reports with equivalent specificity (74%).

Conclusion and/or Teaching Points:

Artificial Intelligence algorithms are not inferior to radiology reports in detecting right heart strain in patients with PE and should play a role in clinical management of PE to expedite thrombectomy in treatment of submassive/massive PE.