First Implementation of an All-in-One Fully Automatic Workflow for Rectal Cancer on an Integrated CT-linac

The study ran a complete integrated sequence of CT simulation, AI autosegmentation, autoplanning, online verification, and beam delivery in 20 rectal cancer patients without repositioning, and the workflow was successfully completed in all 20. These steps are usually distributed across separate devices and stages; this work links the full sequence into a fully automatic chain on a single CT-linac platform. Single-center prospective pilot with 20 patients, using workflow completion as the feasibility endpoint.

Autosegmentation reached high geometric agreement, with mean Dice similarity coefficients of 0.83, 0.95, and 0.89 for gross tumor volume, clinical target volume 45, and clinical target volume 50, respectively. It provides quantitative geometric evidence for the usability of AI autosegmentation on rectal cancer target volumes rather than a qualitative description alone. Evaluated with Dice similarity coefficient and average surface distance in 20 patients.

Autoplanning achieved clinically acceptable target coverage and organ-at-risk sparing in most cases with stable and reproducible plan quality across the cohort; for delivery accuracy, three-dimensional gamma pass rates exceeded 95% in all patients and two-dimensional pass rates ranged from 84.0% to 100.0% across gantry angles under the 3%/3 mm criterion. It evaluates both the dosimetric performance of autoplanning and delivery verification results, indicating that the automated steps are acceptable at the dosimetric level. Based on dosimetric parameters and gamma analysis in 20 patients.

For early clinical outcomes, hematologic toxicity was most common with leukopenia in 55.0% of patients; among 20 patients who completed follow-up, 19 underwent surgery and 1 achieved a clinical complete response, giving an overall complete response rate of 15.0%, with 68.4% of surgical patients reaching tumor regression grade 0-1. Beyond feasibility, it adds preliminary observations of short-term toxicity and surgical pathologic regression, offering early signals for further clinical adoption. Observational follow-up data in 20 patients, at pilot scale.

The results apply to rectal cancer patients receiving radiotherapy in a single-center setting equipped with an integrated CT-linac platform, and describe the feasibility, workflow efficiency, and dosimetric performance of a fully automatic All-in-One workflow; their value lies in providing pilot evidence for further clinical implementation across more centers and cases, and in highlighting autosegmentation and planning as the largest contributors to workflow time, which can serve as targets for later optimization.

As a pilot study, the 20-patient sample and single-center design make the results better suited as a feasibility signal than a general conclusion; autosegmentation and planning account for the largest share of workflow time, and their performance across varying case complexity remains to be observed; two-dimensional gamma pass rates fell to 84.0% at some gantry angles, and the clinical meaning of this warrants continued attention in larger samples; long-term oncologic outcomes and late toxicity are not presented here and would need subsequent follow-up studies.