Tumor phase recognition using cone-beam computed tomography projections and external surrogate information

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Computational Biology

Background:

• Extracting tumor respiratory phase from cone-beam CT (CBCT) projections is challenging due to poor visibility and anatomical obstructions.
• External surrogates for predicting tumor motion have limitations due to potential phase pattern incongruence.
• Accurate tumor phase information is crucial for advanced radiotherapy techniques like 4D-CBCT reconstruction and gating.

Purpose of the Study:

• To develop and validate an algorithm for accurately recovering primary tumor motion oscillation components.
• To combine information from CBCT projections and external surrogates for improved tumor phase prediction.
• To address the limitations of direct CBCT analysis and external surrogate prediction in radiotherapy.

Main Methods:

• A two-step algorithm combining Local Principal Component Analysis (LPCA) on cropped tumor images and Multivariate Singular Spectrum Analysis (MSSA) with external surrogate data.
• Phantom studies using a QUASAR respiratory motion phantom with simulated anatomical obstructions.
• Validation with real-patient breathing patterns and patient studies involving eight patients with various tumor locations.

Main Results:

• The algorithm demonstrated high accuracy in peak and valley detection (-0.009 ± 0.18 sec) with no time delay in phantom studies.
• Robust performance was observed under anatomical obstruction scenarios, with low expiration phase discrepancy (1.6 ± 1.2%).
• Patient studies showed excellent agreement with reference waveforms, achieving -1.05 ± 3.0% overall phase discrepancy.

Conclusions:

• An innovative method accurately recognizes tumor phase information by integrating CBCT projections and external surrogate data.
• The developed algorithm significantly improves prediction accuracy compared to traditional statistical methods.
• This method provides a reliable ground truth for 4D-CBCT reconstruction, treatment gating, and other clinical applications requiring precise tumor phase data.