3D Cardiac Substructures Segmentation from CMRI using Generative Adversarial Network (GAN).

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Disease Diagnosis

Background:

• Cardiac magnetic resonance imaging (CMRI) is crucial for diagnosing cardiovascular diseases, offering high spatio-temporal resolution.
• Accurate segmentation of cardiac substructures is vital for assessing ventricular function (e.g., stroke volume, ejection fraction).
• Manual segmentation is laborious, time-consuming, and prone to errors, necessitating automated solutions.

Purpose of the Study:

• To develop and evaluate a 3D Generative Adversarial Network (GAN) for automated segmentation of cardiac substructures in CMRI.
• To improve the accuracy and efficiency of extracting cardiac function parameters compared to manual methods.
• To leverage 3D contextual information within the GAN for enhanced segmentation performance.

Main Methods:

• Implementation of a 3D Generative Adversarial Network (GAN) incorporating 3D contextual information.
• Evaluation of the 3D GAN on the ACDC dataset, which includes data from four pathologies and one healthy group.
• Performance comparison against existing methods using metrics like Dice score on the ACDC and M&Ms datasets.

Main Results:

• The proposed 3D GAN demonstrated superior segmentation accuracy compared to other methods in the literature.
• The method achieved better performance even when trained with a limited amount of data.
• A higher Dice score was obtained on the blind-tested M&Ms dataset, confirming robust segmentation capabilities.

Conclusions:

• The developed 3D GAN offers an accurate and efficient automated solution for cardiac substructure segmentation in CMRI.
• This AI-driven approach has the potential to significantly aid physicians in diagnosing cardiovascular diseases.
• The method's effectiveness, particularly with limited data, suggests broad applicability in clinical settings.