Dose control in robot-assisted seed brachytherapy: challenges, current status, and future directions

Seed brachytherapy treats a range of localized solid tumors, where efficacy and normal-tissue sparing depend on precise dose control. Advancing it from geometry-controlled execution toward autonomous, dose-controlled delivery is obstructed by the irreversibility of implantation: once released, a seed fixes its dosimetric contribution permanently. Errors arising from plan-to-anatomy mismatch, placement imprecision, and limited intraoperative assessment therefore become embedded in the realized dose distribution, beyond postoperative correction. Correction is available only during the procedure, which places the robot in the path from a computed dose strategy to a verified outcome. Real-time imaging aids visualization, yet the feedback it supports falls short in resolution, in latency, or in its coupling to execution, and does not close the loop. This narrative review maps the technologies for robot-assisted closed-loop dose control onto a five-layer control architecture. The architecture is organized around robotic execution as the physical core of the loop (actuation), together with the layers built upon it: intraoperative sensing and anatomical updating (perception); real-time dose computation (computation); adaptive replanning (decision); formal safety assurance (supervision). Each layer has advanced, though none has reached the threshold for clinical deployment. We identify the bottlenecks and propose a roadmap driven by real-time seed localization resolved for orientation, intraoperative contour recovery, near-real-time dose surrogates, irreversibility-aware replanning, and formally verifiable safety architectures. Beyond brachytherapy, these requirements point toward verifiable autonomy in image-guided interventional surgery, where irreversible actions require robotic execution under provable safety guarantees.