KEY POINTS
- This multicentre technical study included 35 patients across four cohorts: 10 prospective and 25 retrospective cases, comprising 24 thoracic and 11 abdominal cases. Seven patients were undergoing reirradiation.
- The patient-specific MorphTracking framework combined optical surface imaging point clouds with single-view X-ray projections to estimate three-dimensional deformation fields and reconstruct tumors, organs at risk, and surrounding anatomy during respiration.
- Compared with three existing methods, MorphTracking achieved significantly lower reconstruction error and higher peak signal-to-noise ratio across all cohorts (p<0.01). It also produced higher Dice similarity coefficients and lower relative volume errors for tumors and organs at risk (p<0.01).
- During full-cycle evaluation, predicted tumor motion correlated closely with reference motion in the superior–inferior, anterior–posterior, and left–right directions: 0.959 ± 0.029, 0.955 ± 0.036, and 0.923 ± 0.039, respectively. Mean inference time was 15.622 ± 0.001 milliseconds per frame.
- In simulated dynamic multileaf collimator tracking across three reirradiation cases, the framework maintained target-dose indices while reducing spinal cord maximum dose and ipsilateral lung V5 compared with an internal gross tumor volume approach. Physical delivery, end-to-end latency, and clinical outcomes were not evaluated.
CLINICAL TAKEAWAY
The framework could support markerless, voxel-wise tracking of tumors and organs during respiratory motion, potentially enabling more adaptive delivery than rigid external-surrogate tracking. However, the evidence remains computational, and dynamic delivery was simulated in only three cases without physical dosimetry, hardware integration, or clinical validation.
SOURCE
International Journal of Radiation Oncology, Biology, Physics