TC-SegNet: robust deep learning network for fully automatic two-chamber segmentation of two-dimensional...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Accurate heart chamber quantification is crucial for diagnosing cardiac abnormalities.
• Echocardiography-based segmentation requires precise algorithms for reliable analysis.
• Existing methods face challenges in achieving high accuracy in cardiac segmentation.

Purpose of the Study:

• To propose a robust two chamber segmentation network (TC-SegNet) for improved echocardiography analysis.
• To enhance heart chamber segmentation accuracy for better cardiac disease diagnosis.
• To develop a deep learning model leveraging advanced architectural components.

Main Methods:

• Development of TC-SegNet, a U-Net based architecture.
• Integration of modified skip connections, Atrous Spatial Pyramid Pooling (ASPP), and squeeze and excitation modules.
• Evaluation on the Cardiac Acquisitions for Multi-structure Ultrasound Segmentation (CAMUS) dataset.

Main Results:

• TC-SegNet achieved an average F1-score of 0.91, Dice score of 0.9284, and IoU score of 0.8322.
• The model demonstrated a significantly lower Pixel Error (PE) of 1.5109 compared to reference models.
• Quantitative metrics indicate superior performance over existing state-of-the-art segmentation methods.

Conclusions:

• TC-SegNet offers a robust and accurate solution for heart chamber segmentation in echocardiography.
• The proposed model shows significant potential for improving the diagnosis of cardiac diseases.
• The integration of novel modules enhances segmentation performance and outperforms current methods.