Deep learning enables fully automated cineCT-based assessment of regional right ventricular function.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Right ventricular (RV) function is crucial for diagnosing and predicting heart disease.
• Current CT-based RV assessments are manual, time-consuming, and variable.

Purpose of the Study:

• To develop and evaluate a fully automated deep learning pipeline for RV volumetric and regional strain analysis from cineCT images.

Main Methods:

• Developed a two-part deep learning pipeline: RHBS for endocardial boundary segmentation and RVWL for wall labeling.
• Trained and validated models on diverse RV phenotypes and an independent aortic stenosis cohort.

Main Results:

• Achieved high accuracy (Dice scores >0.96) for RV segmentation and volumetry.
• RVWL demonstrated high accuracy (>93%) for wall labeling.
• Combined pipeline accurately assessed regional strain with high cosine similarity (0.97).

Conclusions:

• A fully automated 3D cineCT-based RV regional strain analysis pipeline improves efficiency and reproducibility.
• This approach facilitates large-scale cohort evaluations and multi-center studies.