Detection of left ventricular wall motion abnormalities from volume rendering of 4DCT cardiac angiograms using deep learning

Zhennong Chen; Francisco Contijoch; Gabrielle M Colvert; Ashish Manohar; Andrew M Kahn; Hari K Narayan; Elliot McVeigh

doi:10.3389/fcvm.2022.919751

Detection of left ventricular wall motion abnormalities from volume rendering of 4DCT cardiac angiograms using deep learning

Front Cardiovasc Med. 2022 Jul 28:9:919751. doi: 10.3389/fcvm.2022.919751. eCollection 2022.

Authors

Zhennong Chen¹, Francisco Contijoch^{1

2}, Gabrielle M Colvert¹, Ashish Manohar³, Andrew M Kahn⁴, Hari K Narayan⁵, Elliot McVeigh^{1

2

4}

Affiliations

¹ Department of Bioengineering, UC San Diego School of Engineering, La Jolla, CA, United States.
² Department of Radiology, UC San Diego School of Medicine, La Jolla, CA, United States.
³ Department of Mechanical and Aerospace Engineering, UC San Diego School of Engineering, La Jolla, CA, United States.
⁴ Department of Cardiology, UC San Diego School of Medicine, La Jolla, CA, United States.
⁵ Department of Pediatrics, UC San Diego School of Medicine, La Jolla, CA, United States.

Abstract

Background: The presence of left ventricular (LV) wall motion abnormalities (WMA) is an independent indicator of adverse cardiovascular events in patients with cardiovascular diseases. We develop and evaluate the ability to detect cardiac wall motion abnormalities (WMA) from dynamic volume renderings (VR) of clinical 4D computed tomography (CT) angiograms using a deep learning (DL) framework.

Methods: Three hundred forty-three ECG-gated cardiac 4DCT studies (age: 61 ± 15, 60.1% male) were retrospectively evaluated. Volume-rendering videos of the LV blood pool were generated from 6 different perspectives (i.e., six views corresponding to every 60-degree rotation around the LV long axis); resulting in 2058 unique videos. Ground-truth WMA classification for each video was performed by evaluating the extent of impaired regional shortening visible (measured in the original 4DCT data). DL classification of each video for the presence of WMA was performed by first extracting image features frame-by-frame using a pre-trained Inception network and then evaluating the set of features using a long short-term memory network. Data were split into 60% for 5-fold cross-validation and 40% for testing.

Results: Volume rendering videos represent ~800-fold data compression of the 4DCT volumes. Per-video DL classification performance was high for both cross-validation (accuracy = 93.1%, sensitivity = 90.0% and specificity = 95.1%, κ: 0.86) and testing (90.9, 90.2, and 91.4% respectively, κ: 0.81). Per-study performance was also high (cross-validation: 93.7, 93.5, 93.8%, κ: 0.87; testing: 93.5, 91.9, 94.7%, κ: 0.87). By re-binning per-video results into the 6 regional views of the LV we showed DL was accurate (mean accuracy = 93.1 and 90.9% for cross-validation and testing cohort, respectively) for every region. DL classification strongly agreed (accuracy = 91.0%, κ: 0.81) with expert visual assessment.

Conclusions: Dynamic volume rendering of the LV blood pool combined with DL classification can accurately detect regional WMA from cardiac CT.

Keywords: computed tomography; deep learning; left ventricle (LV); volume rendering (VR); wall motion abnormality detection.

Abstract

Grants and funding