Modeling dose uncertainty in cone-beam computed tomography: Predictive approach for deep learning-based synthetic...

Area of Science:

• Medical Physics
• Radiotherapy
• Medical Imaging

Background:

• Cone-beam computed tomography (CBCT) is crucial for image-guided radiotherapy (RT) but exhibits variable CT numbers compared to standard CT.
• Synthetic CT (sCT) generation from CBCT is necessary for accurate dose calculations but remains challenging due to inherent uncertainties.

Purpose of the Study:

• To develop and validate a voxel-wise uncertainty estimation method for CBCT-to-sCT synthesis.
• To correlate uncertainty maps with sCT-CT errors and quantify dose uncertainties.

Main Methods:

• Developed and validated an uncertainty estimation method using 85 head and neck (H&N) patients treated with photon RT.
• Included three external patients to assess robustness on out-of-distribution images.
• Generated 'plausible' sCTs to explore error scenarios and quantify dose uncertainties.

Main Results:

• Uncertainty maps showed a strong correlation (Pearson's r = 0.65–0.72) with absolute sCT-CT error maps.
• Dose uncertainty was quantified using dose-volume histograms (DVHs).
• Reference CT DVHs were within the uncertainty intervals derived from sCT for most patients.

Conclusions:

• The proposed method effectively predicts uncertainty maps, aiding in sCT quality assessment.
• A novel approach for estimating dose uncertainty is provided by defining confidence intervals around CT DVHs.