Automated biventricular quantification in patients with repaired tetralogy of Fallot using a three-dimensional deep...

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Deep learning models excel at standard cardiac segmentation but struggle with congenital heart defects like Tetralogy of Fallot (TOF).
• Existing models trained on normal or acquired heart disease data are ill-suited for the unique anatomical variations in TOF patients.
• There is a need for specialized deep learning models to accurately assess cardiac function in TOF survivors.

Purpose of the Study:

• To develop and validate a dedicated deep learning model for precise left ventricle (LV) and right ventricle (RV) segmentation and quantification in repaired TOF patients.
• To improve the accuracy of cardiac cavity and myocardium measurements in individuals with congenital heart disease.
• To assess the performance of the developed model against commercial software for clinical utility.

Main Methods:

• A 3D convolutional neural network (CNN) was trained using a 5-fold cross-validation approach on a mixed dataset of normal/acquired cardiac pathology and TOF patients.
• The model incorporated flexibility to handle missing labels for end-diastolic (ED) and end-systolic (ES) phases, common in TOF.
• The best model was applied to a TOF test set, with automated ED/ES phase determination and comparison against commercial software (suiteHEART®).

Main Results:

• Training on a mixed dataset improved performance, achieving high Dice similarity coefficients for LV (93.8% ED, 89.8% ES) and RV (92.9% ED, 90.9% ES) cavities.
• Accurate segmentation of LV (80.9%) and RV (61.8%) myocardium was achieved at ED.
• The model demonstrated comparable LV quantification to commercial software but significantly outperformed it in RV cavity quantification (12 ml vs. 36 ml mean absolute error).

Conclusions:

• A fully automatic deep learning approach for LV and RV segmentation and quantification, including RV mass, was successfully developed and validated for repaired TOF patients.
• The developed model offers superior RV quantification accuracy compared to existing commercial software.
• This approach holds significant potential for improving clinical management and patient care in individuals with repaired TOF.