Back to 2025 Karmody Posters
Video-based AI For Assessment Of Cardiac Function And Aortic Compliance After Aortic Repair
Eileen Lu, MD, Justin Rhee, Shruthi Nammalwar, Donald T. Baril, Navyash Gupta, Cassra N. Arbabi, Ali Azizzadeh, David Ouyang, Elizabeth L. Chou.
Cedars-Sinai Medical Center, Los Angeles, CA, USA.
Objective Endografts for thoracic endovascular aortic repair (TEVAR) are much less compliant than the native aorta and have been associated with changes in flow hemodynamics and cardiac remodeling. There is high variance and limited precision with evaluating central hemodynamics on echocardiograms among operators. Video-based deep learning algorithms use segmentation and spatiotemporal convolutions to obtain beat to beat cumulative evaluations of aortic distensibility and strain, which can accurately assess cardiac function and aortic compliance to inform clinical decision making.
Methods 51730 PLAX echocardiogram videos obtained from Cedars-Sinai Medical Center were divided into training (46188), validation (5035), and test (507) sets. Aortic root strain was obtained by dividing the change in aortic root diameter by the minimum aortic root diameter during each cardiac cycle as measured by the deep learning model. An exploratory data analysis was performed on a cohort of 33 patients who underwent endovascular repair to reveal trends between surgical measures and aortic root strain.
Results At a frame-level analysis, the deep learning model accurately measured aortic root diameter with a mean absolute error (MAE) of 2.5mm. 33 patients underwent endovascular repair with mean age 64.9 +/- 11.8 and BMI 27.4 +/- 6.6. There were 15(46%) patients with dissection and 27(84%) patients with aneurysm. Exploratory data analysis revealed increasing aortic root strain after endovascular repair (0.082 vs 0.122, p=0.003), but no directional trend after a combination of endovascular and open repair (0.100 vs 0.103, p=0.812) (Figure 1). Preoperative aortic root strain was lower in patients with prior abdominal surgical history (0.059 vs 0.129, p=0.001) and higher in those on Plavix (0.178 vs 0.091, p=0.006). Postoperative aortic root strain was lower in patients requiring surgery within 30 days (0.076 vs 0.127, p=0.041).
Conclusion The deep learning model accurately measures aortic root diameter and identifies subtle changes in aortic compliance with high accuracy and precision. Exploratory analysis suggests trends in aortic root strain after endovascular repair, which shows promising utility of this model for future clinical application.
Back to 2025 Karmody Posters
