Statistical modeling of 4D respiratory lung motion using diffeomorphic image registration

Area of Science:

• Medical Imaging
• Computational Anatomy
• Radiotherapy Physics

Background:

• Respiratory motion significantly impacts medical imaging accuracy, particularly in lung cancer radiation therapy.
• Current motion modeling is limited to intra-patient registration, restricting broader applications.
• Developing a generalized lung motion model is crucial for advancing thoracic imaging analysis.

Purpose of the Study:

• To develop a statistical 4D mean motion model (4D-MMM) for respiratory lung motion using multi-patient data.
• To enable adaptation of the 4D-MMM to patient-specific lung geometries.
• To evaluate the model's accuracy in predicting lung and tumor motion for clinical applications.

Main Methods:

• Intra-subject registration to create subject-specific motion models.
• Generation of an average lung shape and intensity atlas.
• Registration of subject-specific models to the atlas using symmetric diffeomorphic nonlinear intensity-based registration.
• Log-Euclidean framework for statistical analysis of transformations.
• Adaptation of the 4D-MMM to patient-specific geometries.

Main Results:

• A 4D-MMM was successfully built from 17 patients' thoracic 4D CT data.
• Evaluation on 10 lung cancer patients yielded a mean target registration error (TRE) of 3.3 ±1.6 mm for landmark and tumor motion.
• Prediction accuracy was independent of tumor size and motion amplitude, but decreased with tumors adhering to non-lung structures.
• The model demonstrated potential for radiation therapy and image-guided diagnosis applications.

Conclusions:

• The developed statistical 4D mean motion model effectively captures average respiratory lung motion.
• The model offers valuable prior knowledge for improving accuracy in thoracic medical imaging applications.
• Further research can explore its integration into clinical workflows for enhanced treatment planning and diagnosis.