P2TC: A Lightweight Pyramid Pooling Transformer-CNN Network for Accurate 3D Whole Heart Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Research

Background:

• Cardiovascular disease is a major cause of mortality globally.
• Accurate 3D whole heart segmentation is crucial for diagnosis and surgical planning.
• Existing deep learning methods face challenges in spatial context, long-range dependencies, and local feature representation for 3D segmentation.

Purpose of the Study:

• To develop a novel, lightweight network for accurate 3D whole heart segmentation.
• To address limitations in current deep learning models for cardiac segmentation.
• To enhance the precision and efficiency of segmenting cardiac structures in medical images.

Main Methods:

• Proposed a Pyramid Pooling Transformer-CNN (P2TC) network with a dual encoder-decoder structure.
• Integrated a 3D pyramid pooling Transformer for multi-scale information fusion.
• Employed a lightweight large-kernel Convolutional Neural Network (CNN) for local feature extraction and a two-branch decoder for precise segmentation and contextual residual handling.

Main Results:

• P2TC achieved state-of-the-art performance on the MM-WHS 2017 challenge dataset.
• Achieved Dice scores of 92.6% for Computed Tomography (CT) and 88.1% for Magnetic Resonance Imaging (MRI).
• Outperformed the baseline model by 1.5% (CT) and 1.7% (MRI), demonstrating superior segmentation accuracy.

Conclusions:

• The proposed P2TC network offers a significant advancement in 3D whole heart segmentation.
• P2TC effectively models multi-scale and local features, overcoming limitations of previous methods.
• The model demonstrates high accuracy and efficiency for cardiac segmentation across different imaging modalities.