A Deep Learning-Based Fully Automated Cardiac MRI Segmentation Approach for Tetralogy of Fallot Patients

Area of Science:

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

Background:

• Automated cardiac MRI segmentation offers improved accuracy and reproducibility for ventricular function assessment in Tetralogy of Fallot (ToF).
• Manual segmentation is time-consuming and prone to variability, highlighting the need for automated solutions.

Purpose of the Study:

• To evaluate deep learning (DL) models for automatic segmentation of the left ventricle (LV), right ventricle (RV), and LV myocardium in ToF patients.
• To compare DL model performance against manual reference standard annotations.

Main Methods:

• A retrospective study involving 427 patients (122 ToF) with cardiac conditions, using cardiac MRI (1.5/3T cine SSFP).
• U-Net, Deep U-Net, and MultiResUNet models were trained on diverse datasets and evaluated using Dice Similarity Coefficient (DSC), Intersection over Union (IoU), and F1-score.
• Statistical analysis included Friedman tests, Wilcoxon tests, and Pearson's correlation to assess segmentation accuracy and functional parameter agreement.

Main Results:

• The MultiResUNet model trained on a mixed dataset achieved the highest segmentation performance, with DSCs of 96.1% for LV and 93.5% for RV.
• Internal testing demonstrated high DSCs (e.g., 97.3% for LV, 94.7% for RV at end-diastole) and strong correlations (0.84-0.99) for ventricular function measurements.
• External validation confirmed generalizability, with correlations for ventricular measurements ranging from 0.81 to 0.98.

Conclusions:

• The MultiResUNet model provides accurate automated cardiac MRI segmentation for Tetralogy of Fallot.
• This automated approach has the potential to streamline clinical workflows and enhance disease monitoring in ToF patients.