Robust Real-Time Cancer Tracking via Dual-Panel X-Ray Images for Precision Radiotherapy.

Area of Science:

• Medical Physics
• Artificial Intelligence in Medicine
• Radiotherapy Technology

Background:

• Respiratory motion significantly challenges lung cancer radiotherapy precision.
• Accurate tumor localization is crucial for effective radiation delivery.
• Current tracking methods often rely on markers or have limitations in real-time performance.

Purpose of the Study:

• To develop and evaluate a deep learning-based, markerless system for real-time lung tumor tracking using orthogonal X-ray projections.
• To enhance the accuracy and robustness of tumor localization during radiotherapy.
• To assess the system's performance against existing algorithms using diverse datasets.

Main Methods:

• A novel deep learning framework combining advanced data augmentation (hybrid deformable model, 3D thin-plate spline), a Transformer-based segmentation network, and a CNN regression network.
• Utilized orthogonal X-ray projection images for 3D tumor position estimation.
• Evaluated using The Cancer Imaging Archive (TCIA) patient data and dynamic thorax phantom data under varying noise conditions.

Main Results:

• Achieved high segmentation accuracy (DSC ~0.97) and precise 3D centroid localization (deviation < 0.55 mm on patient data, < 0.33 mm on phantom data).
• Demonstrated excellent performance across different tumor sizes and noise levels, with errors < 1 mm even under high noise.
• Real-time processing achieved with an average speed of 90 ms/frame, maintaining high tracking success rates.

Conclusions:

• The proposed markerless tracking method offers superior accuracy, noise robustness, and real-time capabilities for lung tumor localization in radiotherapy.
• This technology has the potential to significantly improve radiotherapy precision and personalize cancer treatment, particularly for small tumors.
• Represents a significant advancement in overcoming respiratory motion artifacts in lung cancer treatment.