Template-guided reconstruction of pulmonary segments with neural implicit functions

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• Accurate 3D reconstruction of pulmonary segments is vital for lung cancer surgery and planning.
• Conventional deep learning methods face limitations in resolution and computational resources.
• Implicit modeling offers computational efficiency and continuous representation for 3D shapes.

Purpose of the Study:

• To develop an anatomy-aware, precise 3D pulmonary segment reconstruction method using neural implicit functions.
• To introduce clinically relevant metrics for evaluating reconstruction quality.
• To create a benchmark dataset (Lung3D) for pulmonary segment reconstruction algorithms.

Main Methods:

• A neural implicit function-based approach to learn 3D surfaces for pulmonary segment reconstruction.
• Deformation of a learnable template to represent the pulmonary segment shape.
• Development of two novel clinical evaluation metrics for reconstruction assessment.
• Creation of the Lung3D dataset with 800 labeled pulmonary segments and associated vasculature.

Main Results:

• The proposed neural implicit function method achieves precise 3D pulmonary segment reconstruction.
• The method demonstrates superior performance compared to existing reconstruction techniques.
• The Lung3D dataset provides a valuable resource for benchmarking and advancing pulmonary reconstruction algorithms.

Conclusions:

• The neural implicit function-based method offers a significant advancement in 3D pulmonary segment reconstruction.
• This approach enhances anatomical accuracy and detail, benefiting surgical planning.
• The study provides a new perspective and valuable resources for the field of pulmonary reconstruction.