Generalizable Framework for Atrial Volume Estimation for Cardiac CT Images Using Deep Learning With Quality Control...

Area of Science:

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

Background:

• Cardiac computed tomography (CCT) is crucial for pre-operative assessment in atrial fibrillation (AF) catheter ablation (CA).
• Left atrial volume (LAV) measurement from CCT aids in stratifying AF recurrence risk, as larger volumes correlate with higher recurrence rates.
• Current LAV measurement methods can be time-consuming and may lack standardization.

Purpose of the Study:

• To develop and validate a deep learning (DL) framework for automated computation of left atrial volume (LAV) from CCT images.
• To assess the accuracy and quality control of the automated LAV estimation process.
• To enable high-throughput and reproducible LAV measurements for clinical decision-making.

Main Methods:

• A framework utilizing ResNet50 for image classification and UNet for image segmentation was developed.
• The framework processes CCT images to identify, segment, and estimate the volume of the left atrium (LA).
• Quality control mechanisms were integrated to assess segmentation accuracy and identify poor LAV estimations.

Main Results:

• The DL framework achieved 98% accuracy in image classification and an 88.5% mean dice score for LA segmentation.
• Segmentation quality prediction reached an 82% mean dice score, with an R-squared value of 0.968 for volume estimation.
• The system correctly identified 90% of poor LAV estimations, demonstrating robust quality control.

Conclusions:

• A generalizable DL framework for automated LAV estimation from CCT was successfully developed.
• The framework offers an efficient and robust method for quality assessment of DL-based image segmentation and volume estimation.
• This approach facilitates high-throughput extraction of reproducible LAV measurements, supporting clinical applications in AF management.