Dilated-Inception Net: Multi-Scale Feature Aggregation for Cardiac Right Ventricle Segmentation

Area of Science:

• Medical Imaging
• Cardiology
• Artificial Intelligence

Background:

• Cardiac ventricle segmentation is crucial for diagnosing and monitoring heart conditions.
• Manual segmentation is time-consuming and subjective, necessitating automated methods.
• Existing automated methods struggle with the complex anatomy of the right ventricle.

Purpose of the Study:

• To develop an effective automated segmentation model for the right ventricle.
• To address the challenges posed by the right ventricle's complex shape and ambiguous boundaries.
• To achieve expert-level performance in right ventricle segmentation using deep learning.

Main Methods:

• Proposed a novel Dilated-Inception Net (DIN) for feature extraction and aggregation.
• Utilized multi-scale feature extraction to handle complex anatomical variations.
• Trained and evaluated the model on a benchmark dataset for right ventricle segmentation.

Main Results:

• The proposed DIN model outperformed state-of-the-art methods on the benchmark dataset.
• The model demonstrated potential for expert-level performance in right ventricular epicardium segmentation.
• DIN showed high correlation with clinical expert assessments in four cardiac indices.

Conclusions:

• The Dilated-Inception Net (DIN) is a promising deep learning model for automated right ventricle segmentation.
• DIN offers a reliable and efficient tool for clinical applications in cardiac imaging.
• The model's performance suggests its utility in improving cardiac disease diagnosis and patient monitoring.