Navigation-Assisted Surgery for Locally Advanced Primary and Recurrent Rectal Cancer

The NAVI-LARRC (Computer navigation-assisted surgery for locally advanced and recurrent rectal cancer) trial was a prospective single-arm trial where the aim was to determine the feasibility of navigation-assisted surgery for LARC and LRRC (NCT 04512937). Surgery was planned and performed using the Brainlab^TM navigation system with Elements software and the Kick® Navigation Station (Brainlab, Munich, Germany) with navigation based on optical tracking with infrared light. Feasibility of navigation was assessed through a tripartite primary endpoint: the R0 resection rate, assessment of surgeons’ opinions on the navigation procedure through a study-specific questionnaire and individual interviews, and determination of adherence to the preoperative resection plan during surgery using postoperative magnetic resonance imaging (MRI). Study approval was obtained from the Regional Ethics Committee of South-East Norway (Approval ID 123753), and written informed consent was obtained from all patients.

All patients were treated at the Norwegian Radium Hospital, a tertiary referral centre for LARC and LRRC requiring surgery beyond TME, and part of Oslo University Hospital Comprehensive Cancer Centre. The study team consisted of two rectal cancer MRI radiologists, five colorectal and three orthopaedic surgeons performing resections, plastic and urologic surgeons providing reconstructive surgery according to need, and two pathologists with experience in evaluation of multivisceral specimens. Routine staging of patients included clinical examination under anaesthesia, computer tomography (CT) of the thorax, abdomen and pelvis, and pelvic MRI to determine tumour stage for LARC,¹¹ and the location, size and number of lesions in LRRC cases. Inclusion criteria were LARC and LRRC with high risk of R1/R2 resection, where the multidisciplinary team (MDT) deemed that navigation-assisted surgery would be likely to improve the probability of obtaining R0 resection. Tumours were additionally classified according to the MRI compartmental classification of rectal cancer suggested by the Royal Marsden group, with involvement defined as tumour extending to (≤ 1 mm), or infiltrating structures.¹² Thirty-day complications were classified according to the Accordion classification system.¹³ The first follow-up at 3 months consisted of clinical evaluation, a CT of the thorax, abdomen and pelvis, and the study-specific pelvic MRI to determine whether the resection plan had been followed.

Preoperative planning was based on 2-dimensional (2D) MRI and a 3-dimensional (3D) virtual CT-derived model, showing the spatial relations between tumour and adjacent structures (Fig. 1 and Supplementary Video 1). To create the 3D model, the study radiologists first outlined tumour boundaries on 2D MRI (1.5 or 3 Tesla, 1-mm axial T2 weighted high-resolution slices) taken after neoadjuvant treatment. The MRI images were subsequently imported into the Brainlab^TM Elements platform where segmentation, i.e. delineation of tumour boundaries and key adjacent anatomical structures, was performed by a surgeon (AMS), in accordance with the radiologists’ assessment. MRI was thereafter fused with CT (soft tissue kernel, slice thickness ≤ 1 mm), whereby segmented anatomy was transposed to axial, sagittal and coronal CT images, and to a 3D model of the pelvic bone automatically segmented from the CT. Prior to surgery, the entire surgical team evaluated this model together with the conventional 2D MRI images, reaching consensus on a resection plan for each patient.

Intraoperative navigation was performed by visualising the surgical instruments in real-time on the preoperative CT images and in the 3D model. To achieve this, registration was performed, where the preoperative images were matched to the patient at surgery (Fig. 2 and Supplementary Video 2). With the patient in the supine position, a patient tracker was first fixed to the anterior iliac crest with two 5-mm threaded pins, ensuring a stable position relative to the pelvic bone. Intraoperative imaging of the patient’s pelvis was thereafter done with a 3D C-arm fluoroscopy unit (Ziehm Vision FD Vario 3D, Ziehm Imaging^TM, Nürnberg, Germany). During image acquisition, a camera emitting infrared light was used for optical tracking of reflective spheres on the patient tracker and the fluoroscopy C-arm, allowing image data of pelvic bone from fluoroscopy to be positioned relative to the patient tracker. 3D fluoroscopy images were converted to 2D fluoroscopy images and subsequently matched with the preoperative CT images in the axial, sagittal, and coronal plane, thereby positioning also the preoperative CT images and the 3D model relative to the patient tracker. Subsequent simultaneous optical tracking of the patient tracker and reflective spheres on the surgical instruments allowed the computer to show the exact position of surgical instruments in real-time on axial, sagittal and coronal CT images and in the 3D model (Fig. 3 and Supplementary Video 3). The accuracy of the registration was determined in the cranio-caudal, antero-posterior and medio-lateral planes by placing the tip of a navigated pointer on exposed bony landmarks (i.e. pubic symphysis, sacral promontory, and iliac crest) to measure the mismatch (in mm) between the actual position in the patient and the position on the navigation screen. All navigation procedures were performed with the patient in the supine position.

After each procedure, participating surgeons completed the study-specific questionnaire containing seven statements with response options graded on a 5-point Likert scale (Fig. 4).¹⁴ The questionnaire focused on the value of segmentation and 3D imaging and explored the use and significance of navigation. The surgeons were also asked to specify structures that had been identified by navigation during surgery. After 12 procedures, a qualitative researcher (LF) conducted individual semi-structured interviews with the eight study surgeons, focusing on the intraoperative use of the navigation technology. Interviews were transcribed verbatim and analysed with a reflexive thematic approach, where inductive coding and division into categories and themes were done (by LF and AMS) to identify patterns of meaning.¹⁵ Written informed consent was given by all participating surgeons.

The surgical specimen was marked by the surgeon to identify anatomical structures and areas at risk of involved margins. In addition, the study pathologists were provided with a form indicating structures involved by tumour according to MRI. Histopathological examination was performed according to the recently published PelvEx Collaborative guidelines for pelvic exenterative practice.¹⁶ During macroscopic examination, the specimen slices were photographed and the images annotated for identification of specific areas of interest. Primary tumours were classified according to TNM 8.¹¹ Response to neoadjuvant therapy was assessed using the four-tier system for tumour regression score (TRS) suggested by the College of American Pathologists.¹⁷ Circumferential resection margins (CRM) were classified as R0 (> 1 mm between the tumour and the resection margin), R1 (≤ 1 mm), or R2 (local macroscopic residual tumour) for both LARC and LRRC.¹⁸

In patients with R1 resection, the specific location of margin involvement was annotated on photomicrographs, and on the macroscopic images of specimen slices. The corresponding preoperative MRI slice containing the segmented tumour was identified to visualise the structures involved in the R1 resections (Supplementary Fig. S1). This allowed assessment of whether the specific location was included in the original resection plan, and whether R1 resection occurred in navigated areas.

Since the study investigated feasibility, there was no formal inclusion target or power calculation. Results are presented as absolute numbers or median with min-max values unless otherwise stated. All data were prospectively registered.

Out of 226 patients with LARC and 19 with LRRC scheduled for beyond-TME surgery between 1 October 2020 and 1 October 2022, 20 patients were included in the study. Three patients were excluded upon MDT reassessment for the following reasons: navigation deemed unnecessary (n = 1), deferral from curative surgery due to age and comorbidity (n = 1), and irresectable metastatic disease (n = 1). The remaining 17 patients underwent navigation-assisted surgery with curative intent: LARC (n = 8) and LRRC (n = 9, including one patient with sigmoid cancer recurrence requiring sacral resection) (Table 1). All patients received neoadjuvant oncological treatment prior to surgery: chemoradiation (CRT) consisting of 2 Gy x 25 with concomitant capecitabine (n = 8), induction chemotherapy consisting of four cycles of chemotherapy followed by CRT (n = 3), re-irradiation with 1,2 Gy x 2 x 17 (n = 3) or 1,5 Gy x 2 x 15 (n = 2), and short-course radiation (5 Gy x 5) followed by consolidation chemotherapy (n = 1). All except one patient had ECOG performance status 0. The median number of compartments involved was 4 (3–5), with the lateral compartment being involved in all patients, and the posterior in 16 of 17 cases.

Operations were performed a median time of 10 weeks (7–20 weeks) after neoadjuvant therapy, and a median 5 weeks (2–14 weeks) after the last MRI. Rectal resections consisted of total pelvic exenterations (TPE, n = 6), abdominoperineal resections (APR, n = 9), and rectal resection with end colostomy (Hartmann, n = 1). Fifteen patients had abdominosacral resections, the level of sacral resection being S3 or below, except in one patient who had high subcortical sacrectomy of S1–S3.¹⁹ Thirteen patients had resection of sacral nerves S3 or above, three with complete sciatic nerve resection (Table 2). Fifteen patients had perineal reconstruction with a pedicled vertical rectus abdominis myocutaneous flap. The median blood loss was 1.6 l (0.5–8.5 l) and 4.5 l (2.0–6.1 l), and median time of operation was 11 h (9.1–17.5 h) and 18 h (8.5–22.6 h) for LARC and LRRC, respectively.

The sacral nerve roots were transected at a more proximal level than planned in three patients: the left S1 nerve while controlling bleeding from the superior gluteal vessels (n = 1), the right S1 nerve during dissection of dense fibrotic presacral tissue (n = 1), and the right S2 nerve during resection of the sacrospinal ligament (n = 1). The first two instances occurred prior to initiating navigation, and the third after completing the navigation procedure.

Ten patients experienced Accordion complications grade 3 or above; mechanical ventilation more than 72 h (n = 1), deep surgical site infection requiring radiologic drainage (n = 8), and stenosis of the ureter requiring nephrostomy (n = 1). No patients required reoperation and there was no 30-day mortality, but one patient died 38 days after surgery with the autopsy showing acute pyelonephritis around an indwelling ureteral stent. The median time of hospital stay was 22 days (14–56 days).

The navigation platform was used both in preoperative planning and for intraoperative navigation in all patients (Table 2). For preoperative planning, a median time of 3.5 h (1.5–6.3 h) were spent on segmentation, with a median 6 structures (5–11 structures) being segmented in each patient. At surgery, patient registration was completed in a median time of 42 min (30–73 min), and an accuracy of median 1 mm (0–3 mm) was achieved.

Eight participating surgeons completed a total of 60 study-specific questionnaires (Fig. 4). The surgeons agreed that the virtual 3D model with segmented structures gave a better understanding of the tumour situation than 2D MRI alone, and that navigation guided surgery by confirming or identifying landmarks. When asked whether the surgery could have been conducted just as well without navigation, most surgeons disagreed. Navigated structures were all located in the posterior and lateral compartments (Table 2), and most commonly included the neuroforamina with sacral nerves (n = 17), the level of sacral resection (n = 15), and the sciatic spine (n = 16) (Fig. 5).

In the individual interviews, intraoperative navigation was described as feasible and useful by all surgeons (Supplementary Table S1). Navigation provided increased confidence and resolve during dissection, and four out of eight surgeons thought they saved time by navigating. All surgeons considered that the inadvertent nerve resections could have been avoided with more extensive use of navigation. Seven of eight surgeons reported navigation to be particularly helpful in LRRC because of altered anatomy and fibrosis, resulting in more accurate tumour resection and better protection of vital structures compared with non-navigated surgery.

Intestinal adenocarcinoma was confirmed in all resected specimens, with moderate response to neoadjuvant treatment in most cases (TRS1, n = 2; TRS2, n = 14; TRS3, n = 1) (Supplementary Table S2). R0 resection was achieved in 6/8 LARC cases [CRM median 5 mm (1.5–6 mm)], and in 6/9 LRRC cases [CRM median 3.5 mm (1.5–30 mm)].

The 3-month postoperative MRI performed to evaluate adherence to the preoperative resection plan was available for 16 of the 17 patients (Supplementary Table S3). The postoperative MRI confirmed a high degree of adherence to the preoperative resection plan (165 out of 182 structures were resected as planned) with the following discrepancies: deliberate intraoperative adjustments (n = 8), MRI failing to identify partial resection of the pelvic sidewall fascia (n = 1) and pelvic muscles (n = 3), the inadvertent nerve resections (n = 3), and one patient where the planned resection of the obturator muscle (n = 1) and the sciatic spine was not performed (n = 1). Thus, the postoperative MRI provided little additional information relative to what was already recognised at the time of surgery.

Determining the location of R1 on preoperative MRI showed that these occurred in the vagina and paraprostatic tissue in the two LARC patients, in relation to the sacrospinal ligaments in two LRRC patients, and in fat surrounding a presacral tumour in the last LRRC patient (Supplementary Fig. S1). None of these structures had been segmented for navigation preoperatively, and in the last LRRC case, R1 occurred prior to initiating navigation.