KEY POINTS
- This retrospective, single-center study included 29 patients receiving left-sided whole-breast intensity-modulated radiotherapy to 40.05 Gy in 15 fractions, with 87 weekly cone-beam computed tomography scans available for model development and validation.
- Week 1 and Week 2 cone-beam computed tomography scans were used for patient-specific calibration, while Week 3 imaging served as the test set. Monocular surface images were mapped to three-dimensional deformation fields using hybrid surface features, principal component analysis, and physics-informed regularization.
- The framework achieved a mean gamma passing rate of 93.8% ± 3.1% at 3%/3 mm, a dose correlation of 0.940 ± 0.037, and a mean anatomical displacement error of 1.12 ± 0.53 mm.
- Mean reconstruction latency was 42 ± 5 milliseconds, approximately 124-fold faster than the commercial deformable registration benchmark of 5.2 ± 0.8 seconds.
- Removing physics-based regularization reduced gamma passing rate from 94.2% to 85.7% (p<0.001), while removing principal component reduction lowered it to 72.1% (p<0.001). Validation was limited to one institution, left-sided cases, Week 3 imaging, and dosimetrically defined surrogate structures.
CLINICAL TAKEAWAY
Patient-specific monocular surface imaging could fill the monitoring gap between intermittent volumetric scans by providing rapid, non-ionizing estimates of three-dimensional dose during breast radiotherapy. However, retrospective single-center validation, reliance on surrogate regions of interest, and limited temporal testing mean the system requires prospective multicenter evaluation before it can support beam-hold decisions or adaptive replanning.