Dataset Information

Retraining Convolutional Neural Networks for Specialized Cardiovascular Imaging Tasks: Lessons from Tetralogy of Fallot.

ABSTRACT: Ventricular contouring of cardiac magnetic resonance imaging is the gold standard for volumetric analysis for repaired tetralogy of Fallot (rTOF), but can be time-consuming and subject to variability. A convolutional neural network (CNN) ventricular contouring algorithm was developed to generate contours for mostly structural normal hearts. We aimed to improve this algorithm for use in rTOF and propose a more comprehensive method of evaluating algorithm performance. We evaluated the performance of a ventricular contouring CNN, that was trained on mostly structurally normal hearts, on rTOF patients. We then created an updated CNN by adding rTOF training cases and evaluated the new algorithm's performance generating contours for both the left and right ventricles (LV and RV) on new testing data. Algorithm performance was evaluated with spatial metrics (Dice Similarity Coefficient (DSC), Hausdorff distance, and average Hausdorff distance) and volumetric comparisons (e.g., differences in RV volumes). The original Mostly Structurally Normal (MSN) algorithm was better at contouring the LV than the RV in patients with rTOF. After retraining the algorithm, the new MSN + rTOF algorithm showed improvements for LV epicardial and RV endocardial contours on testing data to which it was naïve (N = 30; e.g., DSC 0.883 vs. 0.905 for LV epicardium at end diastole, p < 0.0001) and improvements in RV end-diastolic volumetrics (median %error 8.1 vs 11.4, p = 0.0022). Even with a small number of cases, CNN-based contouring for rTOF can be improved. This work should be extended to other forms of congenital heart disease with more extreme structural abnormalities. Aspects of this work have already been implemented in clinical practice, representing rapid clinical translation. The combined use of both spatial and volumetric comparisons yielded insights into algorithm errors.

SUBMITTER: Tandon A

PROVIDER: S-EPMC7990832 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:ImportanceAdults late after total correction of tetralogy of Fallot (TOF) are at risk for major complications. Cardiovascular magnetic resonance (CMR) imaging is recommended to quantify right ventricular (RV) and left ventricular (LV) function. However, a commonly used risk model by Khairy et al requires invasive investigations and lacks CMR imaging to identify high-risk patients.ObjectiveTo implement CMR imaging in noninvasive risk stratification to predict major adverse clinical outcomes.Design, setting, and participantsThis multicenter study included 575 adult patients with TOF (4.083 patient-years at risk) from a prospective nationwide registry in whom CMR was performed. This study involved 5 tertiary referral centers with a specialized adult congenital heart disease unit. Multivariable Cox hazards regression analysis was performed to determine factors associated with the primary end point. The CMR variables were combined with the noninvasive components of the Khairy et al risk model, and the C statistic of the final noninvasive risk model was determined using bootstrap sampling. The data analysis was conducted from January to December 2016.Main outcomes and measuresThe composite primary outcome was defined as all-cause mortality or ventricular arrhythmia, defined as aborted cardiac arrest or documented ventricular fibrillation and ventricular tachycardia (lasting ≥30 seconds or recurrent symptomatic).ResultsOf the 575 patients with TOF, 57% were male, and the mean (SD) age was 31 (11) years. During a mean (SD) follow-up of 7.1 (3.5) years, the primary composite end point occurred in 35 patients, including all-cause mortality in 13 patients. Mean (SD) RV ejection fraction (EF) was 44% (10%), and mean (SD) LV EF was 53% (8%). There was a correlation between RV EF and LV EF (R, 0.36; 95% CI, 0.29-0.44; P < .001). Optimal thresholds for ventricular function (RV EF <30%: hazard ratio, 3.90; 95% CI, 1.84-8.26; P < .001 and LV EF <45%: hazard ratio, 3.23; 95% CI, 1.57-6.65; P = .001) were independently predictive in multivariable analysis. Both thresholds were included in a point-based noninvasive risk model (C statistic, 0.75; 95% CI, 0.63-0.85) and combined with the noninvasive components of the Khairy et al risk model.Conclusions and relevanceIn patients with repaired TOF, biventricular dysfunction on CMR imaging was associated with major adverse clinical outcomes. The quantified thresholds (RV EF <30% and LV EF <45%) may be implemented in noninvasive risk stratification.

Dataset Information

Retraining Convolutional Neural Networks for Specialized Cardiovascular Imaging Tasks: Lessons from Tetralogy of Fallot.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets