4D-Precise: Learning-based 3D motion estimation and high temporal resolution 4DCT reconstruction from treatment 2D+t...

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Science

Background:

• Respiration-induced motion in radiotherapy introduces uncertainty, potentially leading to dose delivery errors.
• Conventional 4D-CBCT suffers from artifacts, inaccurate motion characterization, and post-hoc limitations.
• Accurate 4D imaging is crucial for mitigating motion-related uncertainties in radiotherapy.

Purpose of the Study:

• To develop a deep-learning motion model for estimating 3D+t CT images from treatment kV projection series.
• To overcome the limitations of conventional 4D-CBCT in capturing breathing motion.
• To enable real-time or near-real-time 4D imaging for improved radiotherapy guidance.

Main Methods:

• Proposed an end-to-end learning-based 3D motion modeling and 4DCT reconstruction model named 4D-Precise.
• Developed a Torch-DRR module for end-to-end training with Digitally Reconstructed Radiographs (DRRs).
• Introduced a novel loss function to regulate spatio-temporal motion field variations, leveraging planning 4DCT for prior motion distribution.

Main Results:

• 4D-Precise was trained patient-specifically and validated on simulated and real patient data.
• Reconstructed volumes closely resembled ground-truth volumes, showing high similarity and accuracy.
• The model achieved smoother deformations and fewer negative Jacobian determinants compared to existing methods like SuPReMo.

Conclusions:

• The proposed 4D-Precise model computes both intra-cycle and inter-cycle breathing motions, unlike conventional techniques.
• It represents motion over an extended timeframe, covering several minutes of kV scan series.
• This advancement offers improved motion management for radiotherapy treatment planning and delivery.