# Computed tomography−based prediction of commissural positions facilitates valve-sparing aortic root replacement

## TL;DR

Using CT scans and deep learning to plan aortic root surgery improves accuracy and reduces intraoperative adjustments compared to traditional methods.

## Contribution

A deep-learning algorithm accurately predicts commissural positions for valve-sparing aortic root replacement using CT scans.

## Key findings

- CT-based planning reduced aortic crossclamp time and intraoperative adjustments compared to conventional methods.
- Algorithm-based CT measurements showed excellent agreement with expert measurements for key surgical parameters.
- CT planning led to fewer cusp repairs and more accurate commissural positioning during surgery.

## Abstract

This study aimed to compare aortic root reimplantation planning using computed tomography (CT) with conventional methods and to assess the accuracy of automated CT measurements using a deep-learning algorithm.

Twenty patients underwent David reimplantation at our hospital with CT-based planning to determine graft sizes from virtual basal ring dimensions and predict commissural positions using electrocardiogram-gated CT. In the controls (n = 20), preoperative transesophageal echocardiography determined virtual basal ring sizing, whereas surgeons intraoperatively assessed commissural positions. We also analyzed correlations between CT measurements obtained manually by experts and automatically by our deep-learning-based algorithm using 50 cases indicated for David reimplantation.

The CT group had a shorter aortic crossclamp time (P = .001). Horizontal shifting of commissures (80% vs 40%) and uneven commissural heights (95% vs 60%) were more frequently observed in the CT group than in the controls (P = .010) during David reimplantation. The CT group had fewer patients (25% vs 85%) and commissures (15% vs 63%) requiring intraoperative commissural position adjustments than the controls (P < .001). Cusp repair was required in 1 patient in the CT group and 7 patients in the control group (P = .018). Algorithm-based measurements showed excellent agreement with expert measurements for virtual basal ring diameters, intercommissural distances, and commissural heights (differences: −0.3 to 0.1 mm; intraclass correlation coefficients: 0.87-0.97).

For David reimplantation, CT-based planning of graft sizes and commissural positions may reduce the need for intraoperative commissural position adjustments and cusp repair. Our deep learning−based algorithm can replace expert measurements in standardized surgical planning.

Pictorial summary of the methods, results, and implications of this study.