Autoadaptive motion modelling for MR-based respiratory motion estimation.

Area of Science:

• Medical Imaging
• Image-Guided Interventions
• Computational Anatomy

Background:

• Respiratory motion significantly challenges image-guided interventions, causing misalignments and affecting intra-procedural imaging.
• Existing patient-specific respiratory motion models struggle with changing breathing patterns during interventions, necessitating frequent recalibration.
• Classical motion modeling approaches lack adaptability, leading to inaccurate estimations when breathing patterns shift over time.

Purpose of the Study:

• To develop a novel, auto-adaptive motion modeling methodology for image-guided interventions.
• To enable accurate estimation of patient respiratory motion even when breathing patterns change during treatment.
• To overcome the limitations of classical motion models that require recalibration for altered breathing.

Main Methods:

• Proposed a novel methodology for motion modeling with the ability to automatically adapt to new breathing patterns.
• Utilized 2D MR slices from different positions to build and apply the motion model.
• Implemented an auto-adaptive motion model by extending previous work on manifold alignment.

Main Results:

• On synthetic data, the auto-adaptive motion model showed 21.45% greater accuracy compared to a non-adaptive model after a breathing pattern change.
• Demonstrated the method's ability to maintain motion estimation accuracy on real data despite respiratory baseline drift.
• Proof-of-principle demonstrated on cardiac-gated thoracic data.

Conclusions:

• The proposed auto-adaptive motion model effectively addresses the challenge of changing respiratory patterns in image-guided interventions.
• This novel approach enhances the accuracy and reliability of motion estimation, crucial for emerging treatments.
• Future work may overcome limitations such as update frequency and prediction latency by altering acquisition protocols.