Feasibility of optical stereotactic navigation for rectosigmoid cancer with deep learning-supported 3D modelling

Background: The rate of incomplete resection in the surgical treatment of locally advanced primary rectal cancer and locally recurrent rectosigmoid cancer is high. The use of optical stereotactic navigation during surgery has the potential to reduce this risk. This study evaluates the feasibility of its implementation by assessing navigational accuracy and oncological outcomes.

Methods: In this prospective interventional cohort study, ten patients with either locally advanced cT4bN0–2 rectal cancer or locally recurrent rectosigmoid cancer underwent surgery supported by optical stereotactic navigation at a single tertiary referral centre. Real-time navigation was achieved via a tracked pointer or surgical instrument, using preoperative CT images fused with MRI-based segmentations of the tumour and adjacent structures, partially generated by trained deep learning models. Primary outcomes were navigation accuracy and the rate of R0 resection, defined as a tumour-free margin of ≥1 mm.

Results: Navigation demonstrated sub-millimetric accuracy with a median target registration error of 0.7 mm (IQR 0.5–1.1). An R0 resection was achieved in 7 of 10 patients (70 %). When analysing only planes where navigation was used, tumour-free margins were obtained in 8 of 10 cases (80 %). No navigation-related complications occurred. Median additional preparation time for navigation was 26 min (IQR 22–51). Surgeon satisfaction was consistently high.

Conclusions: The demonstrated feasibility and high accuracy support the potential value of this technique, particularly within a tertiary care setting where further refinement and evaluation can take place.

Trial registration: NL-OMON55156, NTR-NL8567, https://trialsearch.who.int/Trial2.aspx?TrialID=NL-OMON55156.