Lung diaphragm tracking in CBCT images using spatio-temporal MRF

Area of Science:

• Medical Physics
• Image Processing
• Computational Biology

Background:

• Accurate monitoring of intra-fraction tumor motion is crucial in thoracic and abdominal radiotherapy (EBRT).
• The lung diaphragm apex is a common internal marker, but its position and shape variations pose tracking challenges.
• Existing methods for tracking diaphragm apex motion on Cone Beam Computed Tomography (CBCT) images require improvement in accuracy and efficiency.

Purpose of the Study:

• To propose an alternative, accurate method for tracking the ipsi-lateral hemidiaphragm apex (IHDA) position on CBCT projection images.
• To develop a computationally efficient algorithm for real-time monitoring of diaphragm motion during radiation therapy.

Main Methods:

• A hierarchical method utilizing a spatio-temporal Markov Random Field (MRF) model to represent diaphragm state.
• Likelihood function derived from 4D-Hough space votes.
• Energy minimization using graph-cuts for optimal state determination.
• Heterogeneous GPU implementation using CUDA for performance acceleration.

Main Results:

• The MRF formulation demonstrated superior accuracy compared to the full search method.
• The GPU-based implementation achieved a 16% performance improvement over existing benchmarks, completing in approximately 25 seconds.
• The method successfully tracked IHDA position on 15 clinical CBCT images.

Conclusions:

• The proposed MRF formulation provides enhanced tracking accuracy by considering all possible combinations within the 4D-Hough space.
• The GPU-accelerated implementation significantly improves computational speed, making the approach viable for clinical use in radiation therapy.
• This method offers a more robust and efficient solution for monitoring intra-fraction tumor motion in EBRT.