A back propagation neural network based respiratory motion modelling method.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Developing accurate respiratory motion models is crucial for lung surgical navigation.
• Existing methods often struggle with the complexities of free breathing, including deep inspiration and expiration.
• Precise tracking of lung tissue movement is essential for minimizing invasiveness and improving outcomes.

Purpose of the Study:

• To develop and validate a backpropagation neural network-based respiratory motion modelling method (BP-RMM).
• To enable precise tracking of arbitrary points within lung tissue throughout the entire respiratory cycle.
• To enhance the accuracy and robustness of lung motion prediction for potential surgical applications.

Main Methods:

• Utilized artificial intelligence algorithms to process internal and external respiratory data from four-dimensional computed tomography (4DCT).
• Employed data augmentation via polynomial interpolation to improve dataset robustness.
• Constructed a backpropagation neural network for comprehensive lung tissue movement tracking.

Main Results:

• The BP-RMM demonstrated significant accuracy in tracking lung tissue motion.
• The average target registration error (TRE) for deep respiration phases was 1.819 mm across 75 marked points.
• TRE for normal respiration phases was substantially lower, with a minimum error of 0.511 mm.

Conclusions:

• The BP-RMM is a highly accurate and robust method for respiratory motion modelling.
• The validated method shows promise as a tool for enhancing surgical navigation within the lung.
• This AI-driven approach offers improved precision for tracking lung dynamics during procedures.