CLASSICA: Validating artificial intelligence in classifying cancer in real time during surgery

Treatment pathways for significant rectal polyps differ depending on the underlying pathology, but pre‐excision profiling is imperfect. It has been demonstrated that differences in fluorescence perfusion signals following injection of indocyanine green (ICG) can be analysed mathematically and, with the assistance of artificial intelligence (AI), used to classify tumours endoscopically as benign or malignant. This study aims to validate this method of characterization across multiple clinical sites regarding its generalizability, usability and accuracy while developing clinical‐grade software to enable it to become a useful method.


BACKG ROU N D
Significant rectal polyps (sessile and >2 cm in size) represent a considerable clinical challenge.While smaller polyps can be addressed by routine endoscopic polypectomy and frank clinical cancer is managed via established paradigms, polyps such as these have the option of undergoing advanced local excision whether by transanal endoscopic resection (transanal minimally invasive surgery [TAMIS] or transanal endoscopic microsurgery) or flexible endoscopic excision (either endoscopic mucosal resection or endoscopic submucosal dissection).
Transanal surgical approaches are preferred over endoscopic mucosal resection or endoscopic submucosal dissection for early stage cancers due to their ability to provide a single complete unfragmented specimen in comparison to other modalities [1,2].
At present it is difficult to characterize significant polyps fully in advance of complete excision.Biopsies may be inaccurate in up to 20% of such lesions [3] and can induce scarring and epithelial displacement that complicates subsequent local excision and pathological analysis.Radiological assessment often over-stages such lesions [4].This is important because, while local excision can be curative for benign and even some earliest-stage cancers, excision of a more advanced cancer may complicate the therapeutic algorithm in the case of a locally advanced rectal cancer requiring neoadjuvant therapy and indeed potentially the technical aspects of any subsequent radical resection [5].Additionally, as effective endoscopic excision of an early cancer requires clear surgical margins, improved tumour characterization preoperatively may impact the choice of endoscopic technique and the dissection plane.Conversely, mistaking a benign lesion for cancer adds time and costs to care, with radical resection bringing unwarranted added risks and functional consequences.Additionally, local excision techniques have relatively high positive margin rates, which risk regrowth in benign lesions [6,7].Therefore, while the technology and training to perform endoscopic approaches have become much more available over the past decade, their major rate-limiting steps are correct patient selection and adequacy of excision.
We have previously demonstrated, through the use of fluorescent indocyanine green (ICG) and near-infrared (NIR) imaging, that perfusion is visibly different between cancerous, benign and healthy tissues [8].This discovery can be exploited for clinical use by the application of artificial intelligence (AI) methods, including computer vision and machine learning.In short, fluorescence intensity is captured over time from multiple video frames, and this intensity is plotted as a curve.As the intensity curves of cancer tissue differ from those of healthy and benign tissue, analysis of the curve features for each pixel across the screen leads to accurate classification (see Figures 1 and 2).Such an AI method has been prototyped previously, including in real time in theatre, demonstrating an accuracy of >80% [8].In this CLASSICA study, we will more broadly clinically validate the basic method of classifying tissue by its fluorescence signal characteristics across multiple centres.
We will also determine the generalizability and usability of the technique while developing clinical-grade software that can automatically perform such classification.The software will then be tested prospectively as a method for visual in situ classification of tissue nature and also as a means for guiding or indeed obviating endoscopic biopsies in comparison to standard, clinical biopsies and as a guide for margination for complete local lesion excision.

The overall trial aims
The primary aim of CLASSICA is to validate the concept of tissue characterization by ICG signal perfusion pattern analysis in the differentiation of benign and malignant rectal polyps and to prove its usefulness alongside standard clinical care in five clinical sites across Europe.This will be achieved by developing and deploying novel software as a medical device, 'CLASSICA-OR' .CLASSICA-OR software draws on a previously tested research algorithm but will advance its basic premise and foundational concepts as a new fully clinical-grade, usable medical device capable of using AI to classify significant rectal polyps and tumours as benign or malignant based on the observed ICG fluorescence pattern.The device will be operational in theatre and work via the operative imagery generated by standard endoscopic surgical cameras.In parallel to the validation of this software, the obtained videos will undergo a thorough mathematical and statistical analysis.This analysis aims to systematically and quantitatively evaluate the extracted data, validating the underlying assumptions and enabling the optimization of the software methods (Figure 3).
The secondary aims of this study, assuming that >80% accuracy in tissue classification is achieved, will be (i) to compare the use of in categorizing rectal cancers according to T stage.The study will also assess this system's real-world usability and its acceptability to patients, surgical teams and clinical services via user surveys and focused study groups (Figure 4).

Trial sites and participating surgeons
CLASSICA is a multicentre prospective study with five participating sites across the European Union (EU) (see Table 1).Lead surgeons at each clinical site have expert experience in the evaluation of rectal polyps and TAMIS procedures.Additional sites may be included as the study progresses.The study will be coordinated by University College Dublin (UCD), who will act as trial sponsor.The study has been funded by Horizon Europe (Project no.101057321), and the protocol is registered with clini caltr ials.gov (identifier NCT05793554).In addition to the clinical sites, this trial will be supported by software, legal, educational and societal partners (Table 3).

Trial population
All patients undergoing transanal local excision for treatment or assessment of a rectal tumour or significant rectal polyp as part of F I G U R E 2 Sample fluorescence intensity curve where red and green represent areas of clinical concern and normal mucosa respectively, as determined by the clinician.These curves demonstrate significant differences in their characteristics, as is typical of cancerous tumours, which was concordant with final pathology.Curves have been normalized to IAP (intensity at peak) value extracted at times of the initial peak of each curve to account for variations in luminance across the field of view in addition to smoothing.

F I G U R E 3 Study workflow at patient level.
standard clinical care are potentially suitable for inclusion in the trial, subject to the inclusion/exclusion criteria shown in Table 2.

Trial design
CLASSICA will be performed in a phased, graded way, moving from initial validation of mathematical models to prospective classification accuracy, AI-guided endoscopic targeted biopsies and lesion margination.The first stage is designed as a prospective multicentre observational study as no new equipment is being utilized, and the dye and imaging systems are being used within their approved licence (i.e., assessment of tissue perfusion).Validation will be assessed based on the sensitivities and specificities achieved by the classification methods already developed on video recordings from the patients' procedure.This dataset will also aid the development of clinical-grade software that will be tested prospectively for accuracy, usability and generalizability.The initial observational nature of this study minimizes deviation from the patients' standard clinical pathway.Additionally, it allows for assessment of the acceptability of the total method (which includes dye administration, NIR observation and video recording) and ease of its integration into the clinical workflow.The usability of the technology will be further assessed by a survey of all users and stakeholders in addition to a technical analysis of video quality.
With proof of initial broadened accuracy in a more diverse group of patients than examined in preliminary work and with proof of initial accuracy of clinical-grade software on retrospectively obtained videos, the study will thereafter evolve into a regulated clinical study of a medical device.The device will be tested as a means for guiding biopsy to the area of the most likely malignant yield in patients in comparison to standard endoscopic biopsies and subsequently as a means to guided margination for patients undergoing local excision.

Sample size
The proposed sample size for this study is 600 patients.This is intended to comprise 400 patients with suspected benign or early stage cancerous rectal polyps considered suitable for transanal resection and an additional 200 patients with known cancerous tumours undergoing transanal assessment in the operating theatre as part of their routine clinical care.Patients undergoing assessment in the operating theatre prior to or after neoadjuvant therapy will also be suitable for inclusion in this group.For validation of this technology, we aim to achieve a sensitivity of 95% and a specificity of 87.5%, consistent  with what was seen in our previous study [8], with power calculations based on such diagnostic sensitivity and specificity (two-sided) (Power Analysis and Sample Size 2021, NCSS, USA) [9].A confidence level (1 − α) of 0.95 and a 1:1 ratio of cancer to benign patients were selected using the Wilson score [10].This resulted in a sample size of

Interventions
Adult patients fitting the inclusion criteria will be identified from referral letters, outpatient clinics, endoscopy lists and multidisciplinary cancer meetings and approached for inclusion in the trial.Patients will undergo standard preoperative work-up, with the overall management of each patient being determined by institutional protocols.

Informed consent
A verbal explanation of the study along with the approved Patient Information Leaflet/Informed Consent Form (see Appendix S1) will be provided by a suitably qualified member of the healthcare team for the patient to consider.The Patient Information Leaflet will provide detailed information about the rationale, design and personal implications of the study.Following the information provision, patients will be given the opportunity to discuss the study with their family and healthcare professionals and will be given time to consider their participation in the study.Ideally, they will be allowed 24 h as a minimum.The right of the patient to refuse consent without given reasons will be respected.Patients will then be formally assessed for eligibility and invited to provide informed, written consent for their participation in the study.Patients may TA B L E 2 Inclusion and exclusion criteria.

Inclusion criteria Exclusion criteria
Patients with a confirmed or suspected rectal polyp/tumour of any size undergoing surgical intervention or assessment OR Patients with a known rectal cancer undergoing surgical intervention or assessment in the operating theatre, including those pre-or postneoadjuvant therapy

Acute or chronic liver impairment
The

TA B L E 3
Non-clinical project partners.
withdraw their participation at any time without giving a reason, and no further data regarding their care will be included.However, data already processed will continue to be used where it has become intricate to the software methods.A copy of the signed consent form will be given to the participant.The original signed form will be retained at the study site.

Baseline measurements
Preoperative work-up including, but not limited to, colonic visualization by either colonoscopy or computerized tomography colonography, staging computerized tomography scan of the chest, abdomen and pelvis, magnetic resonance imaging of the rectum, and assessment of fitness for surgery will be carried out in accordance with local hospital policies at each clinical site.If not included as a standard preoperative work-up, baseline renal and liver function and human chorionic gonadotropin levels (where indicated) will be checked prior to recruitment to ensure compliance with the exclusion criteria.

Preparation
Initial endoscopic evaluation will be performed with a white light assessment of the bowel segment transanally.An access method (e.g., Lone Star retractor, GelPOINT Path or TEO rectoscope) may be needed to ensure appropriate visualization.Trauma to the area will be avoided to minimize bleeding, as this may disturb the infrared video.This will be followed by systemic administration of ICG at 0.25 mg/kg body weight and an NIR evaluation of the tissue of interest for 5 min to allow both inflow and outflow to be visualized.To maintain uniformity and for the purposes of the computational analysis, the same commercially available NIR endoscope (PINPOINT) will be used in each site.Local anaesthetic agents containing adrenaline should be avoided during the procedure as they typically contain sodium metabisulphite as a preservative which may interact with ICG.
For the tumour characteristic recording, an NIR camera with the capability to simultaneously display white light, infrared and fluorescent images will be used.A high definition recording device should be used to record the videos in MP4 format, with a standardized protocol for video acquisition being followed (see Appendix S1 and Video 1).As the trial and software advances, the video may be streamed directly from the camera stack system to the CLASSICA-OR system where appropriate.Alternatively videos may be stored uploaded via CLASSICA-WEB, a PC-based portal through which data may be uploaded and subjected to classification.
Following these steps, the procedure/care continues in the normal fashion.

Biopsies and tissue analysis
In addition to that needed for standard pathological assessment, some additional samples may be taken to map the fluorescence pattern across the polyp area in select patients with precision.Any such samples or indeed additional biopsies at the discretion of the operating surgeon will be taken after ICG assessment to minimize bleeding obscuring the fluorescence.

Video analysis
Video recording from the intervention will be uploaded to the CLASSICA platform, once it is ensured to be free of any personal institutions, in whose hands it will be anonymous, in accordance with GDPR.

Software analysis
The software used in this study performs three distinct tasks: Combinations of various extracted features and classification algorithms will be trialled to achieve optimal sensitivity and specificity.Details of the preliminary algorithm initially proposed can be found in reference 8.

Sample handling
The majority of samples referred to in the protocol are taken as part of a standard of care pathway with the results accessed by the research team.Research samples, including video recordings for analysis under this protocol, will be processed and stored within the UCD Clinical Research Centre/Centre for Precision Surgery at the Mater Hospital.Only defined researchers working on this project as part of the study team will have access to the samples.Videos and clinical data will be stored for 10 years in a secure location in the UCD Clinical Research Centre.Tissue samples taken for the purposes of the study only will be disposed of following analysis.

Postoperative care
Postoperative management will be conducted as is standard for each clinical site.Adverse outcomes will be assessed during the patient's routine postoperative visit.

Data collection and management
All trial data will be stored and managed through purposebuilt data management systems (namely CLASSICA-WEB and CLASSICA-OR cloud-based systems) designed specifically for the purposes of this trial.

Outcomes
The CLASSICA project will validate the concept of tissue characterization by ICG signal perfusion mathematical methods with the assistance of AI.We aim to achieve an accuracy of >80%, as was demonstrated in the previous developmental study.Accuracy will be defined as the percentage of correct classifications (cancer vs. benign) with additional specificity and sensitivity analysis performed.
Usefulness is relative to standard biopsies, and unassisted margin clearance will also be studied.In addition to validation of the underlying concept, this study will determine the usability, including cost-benefit analysis, of such technology through the deployment of real-time classification technology.

Safety evaluation and reporting of adverse events
Patients will be monitored for potential adverse events intra-operatively, immediately postoperatively and at their follow-up outpatient appointments.Given the excellent safety profile of ICG and the observational nature of this study, this follow-up will comprise clinical review only without the need for invasive investigations.Any adverse events that may occur will be recorded and reported directly to the study sponsor.Appropriate clinical trial insurance is in place at each clinical site.

Ethical considerations
This trial will be conducted in accordance with the World Medical

DISCUSS ION
The potential and ingrowth of AI in medicine is expanding rapidly, particularly with regard to diagnostics in image-dominant fields such as radiology, ophthalmology and dermatology [11][12][13].However, its use as a surgical decision-making support tool is less clear [14] and, indeed, given the short timeframes and irreversibility of decisions in surgery, its implementation needs great care.While some patients have reservations about the use of autonomous AI in their care, many are supportive of its use to aid a human clinician in decision-making capacities [15][16][17].Surgeons, too, have been found to have mixed views on the technology, with the greatest consensus being found in the use of AI as a decision support tool rather than an autonomous decision-maker [18,19].Any potential AI solution or support for intra-operative decision-making clearly needs careful evaluation and validation.
Given the current uncertainty and variability in best care regarding significant rectal polyps and other subgroups of rectal cancer (including down-staged rectal cancers) [20], a multifaceted approach to developing better care processes in surgery is needed.AI may play a role in establishing this.This may be achieved through improved classification methods at the time of endoscopy, standardized management pathways following the detection of a significant rectal polyp and optimal operative technique for those cases undergoing surgical intervention.While significant advances have been made in minimally invasive surgical approaches [6,7] and management pathways [21], appropriate classification for significant rectal polyps still lags behind.While several systems have been devised based on size, location and morphology via white light and chromoendoscopy [22][23][24][25][26], these systems have typically been validated for polyps measuring <2 cm, limiting their application to larger polyps.Additionally, these classification methods are complex and somewhat subjective, relying on the clinical experience and expertise of the endoscopist.This is associated with a learning curve period and is subject to inter-endoscopist and intra-endoscopist variability [27].This makes the application of such methods difficult to apply by non-gastroenterologists or those working outside reference centres where large numbers of large polyps are commonly seen.The SPECC programme was established in 2015 in part to address concerns such as this [21].This programme represents an educational endeavour focused on the standardization of management of significant rectal polyps with a focus on appropriate documentation, avoidance of unnecessary biopsy and referral to specialist centres for endoscopic or minimally invasive excision when appropriate.Additional efforts have focused predominantly on technology and technical training for the performance of transanal operations and have been well documented elsewhere [28][29][30].Despite these advances, the issue of unanticipated malignancy within a polyp previously thought to be benign still occurs all too often, and further efforts are warranted to improve diagnostic accuracy without compromising future excision.
As such efforts ideally need to be usable at scale and not just by experts, image analysis methods, including AI, are attractive in concept.Indeed, commercially available AI methods have already been developed for polyp identification, but these still leave characterization to the endoscopist observer and current, imperfect practice protocols [31].
ICG perfusion angiography is used frequently in colorectal surgery in the assessment of bowel perfusion while performing an anastomosis [32,33].Perfusion of cancerous tumours is known to differ from benign tissue due to alterations in the microvascular architecture and extracellular matrix of tumours [34] and we have previously demonstrated that these differences can be revealed via ICG fluorescence NIR imaging analysis with the assistance of AI [8].
While there is interest and effort in developing AI methods for ICG interpretation in perfusion analysis, no robust system yet exists to do so, and clinical validation linked to anastomotic outcomes is difficult.In contrast, the binary outcomes (i.e., cancer or not cancer) initially used in our preliminary work provide a way to link prediction with an important outcome in a shorter timeframe.The methods (including image stabilization, tracking, correlated intensity profiling and analysis and display of results as heatmaps) will have utility for anastomotic and other tissue perfusion analysis.However, given the early developmental nature of this study, the results must be replicated more broadly to confirm validity.This is especially true for AI classification methods, given that errors may occur due to bias or lack of diversity within the training dataset [14,35].The multicentre nature of this trial will allow for a greater diversity of the dataset by including potentially different patient populations and diversity in surgeon and system factors.Furthermore, our prior work on using AI ICG fluorescence analysis to classify rectal tumours has relied on basic computer programming proof of concept and predominantly post hoc analysis.The CLASSICA study aims to develop a bespoke system incorporating this technology into a programme that clinicians may use intra-operatively, ultimately in real-time, while also building a greater clinical experience and including additional rectal cancer surgical experts and sites.
The use of AI in surgery brings up a number of ethical and legal considerations, most especially regarding black-box decision-making [36].Different to white-box learning, in which the manner by which the algorithm outputs have been generated from inputs can be retrieved and explained in mathematical and logical terms, blackbox learning involves relationships between the initial inputs and the resulting outputs that are too complex for humans, even those involved in designing the algorithm, to understand.Experts differ in their opinions on the importance of the interpretability associated with white-box modelling and the superiority of black-box deep learning for complex tasks that may be constrained by an insistence on interpretability or, indeed, by the misunderstanding that may occur as a result of enforced explainability [37,38].These controversies aside, current guidelines typically require that any AI component involved in healthcare or health-related research be transparent and explainable, and upcoming EU legislation is expected to formalize this into law [39,40].The AI method under evaluation in this study is considered transparent and explainable, in line with these current guidelines and proposed future legislation.However, the determination of accuracy, as well as proof of usability of technology such as this, must be further assessed, including operability within the

DATA AVA I L A B I L I T Y S TAT E M E N T
The data generated by this study will be made available in Open Science Framework Depositry at https:// osf.io.
fluorescence to target (or indeed replace) endoscopic biopsies versus conventional endoscopist-guided standard practice, (ii) to delineate margins in patients undergoing transanal endoscopic excision and (iii) to explore the accuracy of fluorescence perfusion analysis F I G U R E 1 Sample image tracking including regions of interest as identified by the clinician.Red areas indicate an area of clinical concern and green indicates normal mucosa for comparison.

F I G U R E 4
Overview of study workflow.TA B L E 1 Participating clinical sites.
340 which was expanded to 400 so as to include an 18% (n = 60) dropout rate (based on our initial experience).As every patient will have endoscopic NIR analysis, the 400 patients target encompasses all the study components.The 200 patients with diagnosed rectal cancer are planned for exploratory investigation as a non-TAMIS cohort based on prior studies, including those following neoadjuvant therapy, to provide comparison statistics to determine whether the same method can have similar clinical value and what impact neoadjuvant therapy may have on the interpretation of fluorescence perfusion patterns.
identifying information and subjected to annotation by the operating surgeon to highlight areas of clinical significance.These videos will then be studied by investigators at UCD and Arctur and the classification software applied and examined for quality and accuracy.The pseudonymized information will also be shared with our software development partners and other collaborating V I D E O 1 Validating AI in classifying cancer in real-time surgery: an instructional video of the CLASSICA technique.
(i) tissue tracking, (ii) curve processing and feature extraction, (iii) classification.Tissue tracking counteracts the impact of fine movements within the video (slight camera movements or gentle respiration) and is achieved through modifications of out-of-the-box tracking algorithms such as CSRT, DIS and DeepFlow (available in OpenCV).Tracking quality estimation is also performed.With the help of the tracking algorithm, the curves are then extracted from each frame for the selected region of interest.Prior to extraction, each frame is processed to remove noise, glare and saturation.The curves also undergo a smoothing process to enhance the clarity of the curve's underlying features, making them more discernible for precise feature extraction and analysis.Tissue classification will be conducted through the use of supervised classification algorithms which are available off the shelf through Python libraries.
Declaration of Helsinki on the Ethical Principles for Medical Research involving Human Subjects.Ethical approval has been provided by the local ethics committee at each clinical site.Participants may withdraw from the study at any point without giving cause.Dissemination Study outcomes will be disseminated among the academic community, patients and the general public.Dissemination will be primarily led by Pintail with support from each study partner.Communication with the academic community will occur via publications in journals, presentations at meetings and collaboration through the European Association for Endoscopic Surgery (EAES) and Institut de Recherche contre les Cancers de l'Appareil Digestif (IRCAD).Patients and the general public will be targeted for dissemination of study results through both traditional and social media outlets in addition to the study webpage (https:// class icapr oject.eu).This is a validation rather than a commercialization/productization study but a successful outcome will enable consideration of clinical deployment via the appropriate regulatory pathway.
clinical theatre environment and acceptability to both patients and surgeons.The partners of this study include Penn State Dickenson Law School and the Centre for Advanced Studies in Biomedical Innovation Law at the University of Copenhagen.As part of the CLASSICA project, researchers at these institutions will explore the current legal and ethical frameworks for AI in surgery and determine where a surgical decision support tool such as CLASSICA-OR fits in this evolving space.Additionally, the outputs of the system will be scrutinized for potential bias or other ethical implications.The rapid acceleration of new surgical technologies and techniques seen in recent decades has led to a renewed focus on the importance of appropriate training in advance of adopting these techniques[41].Although the proposed surgical technique remains unchanged in this study, in terms of transanal endoscopic microsurgeries/TAMIS or other approaches, the additional step of ICG assessment and video acquisition may be new to some surgeons.Crucially, this programme's classification element depends on satisfactory video quality and timing.To optimize this, instructional videos will be supplied alongside the written protocol at the outset of the project (see Appendix S1).These videos have been produced by study partners IRCAD in conjunction with the team at UCD and participating clinical partners.As the study progresses, further educational tools and workshops will be provided to maintain a high standard of consistency.IRCAD has a long history of progressing education, and their partnership on this study allows for the optimization of training both during the study period and, once successful, beyond it.The success of this tool depends not only on clinical validation but also on acceptance within the surgical community.While wellfounded clinical validation and ethical-legal compliance will underpin CE approval of the device, the introduction of this tool into routine clinical practice requires guidance from the relevant surgical societies.EAES, a partner in this study, plays a central role in the development, update and maintenance of surgical clinical guidance documents.The findings of this study will be used to enhance existing guidance for the management of rectal cancers and also inform new guidance relating to the use of AI in a surgical decision support capacity.The CLASSICA study, if successful, will validate the use of ICG fluorescence analysis in the classification of rectal tumours seen previously by achieving a similar accuracy rate in a larger, more diverse patient cohort.Successful implantation of the technology across multiple sites with multiple surgeons will speak to the generalizability and usability of this technology, allowing progression to future studies to assess the role of ICG fluorescence imaging and AI in targeted biopsy and the delineation of margins for TAMIS resections.We aim for CLASSICA-OR to develop into a tool that can be safely utilized in theatres throughout Europe and further afield.CLASSICA is the first AI surgical decision support tool of its kind in Europe, and we hope that it may be used as a framework/benchmark for other studies into the applications of AI in surgery.Ultimately we hope that this will result in a usable clinical tool supporting better patient selection with regard to transanal local excision of rectal tumours, setting a precedent for the use of AI-supported surgical decision-making tools.AUTH O R CO NTR I B UTI O N S Dalli J: Conceptualization; investigation; writing -review and editing; validation.*Arezzo: Investigation; writing -review and editing; validation; supervision.Knol J: Investigation; writing -review and editing; supervision; validation.Rojc J: Software; writing -review and editing.Rodriguez R: Visualization; resources.Moynihan A: Conceptualisation; investigation; methodology; project administration; validation; visualisation; writing -original draft preparation.Hardy N: Conceptualisation; investigation; writing -review and editing; validation.Aigner F: Investigation; writing -review and editing; validation; supervision.Hompes R: Investigation; writing -review and editing; validation; supervision.Tuynman J: Investigation; writing -review and editing; validation; supervision.Cucek J: Software; writingreview and editing.Cahill R: Conceptualisation; funding acquisition; investigation; methodology; project administration; resources; supervision; validation; writing -original draft preparation (supporting), writing -review and editing.ACK N OWLED G EM ENTS Open access funding provided by IReL.FU N D I N G I N FO R M ATI O N This study has been funded by the EU through Horizon Europe, Project no.101057321.Funded by the European Union.Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the Health and Digital Executive Agency.Neither the European Union nor the granting authority can be held responsible for them.CO N FLI C T O F I NTER E S T S TATEM ENT R Cahill receives speaker fees from Stryker Corp, Ethicon/Johnson & Johnson and Olympus, consultancy fees from Arthrex, Medtronic and Distalmotion and holds research funding from Intuitive, the Irish Government: Disruptive Technologies and Innovation Fund in collaboration with IBM Research and Deciphex and the European Union: Horizon Europe in collaboration with Palliare, Steripak and Arctur.J Dalli receives funding from the Tertiary Education Scholarship Scheme, Malta (EU).A Moynihan, N Hardy, F Aigner, A Arezzo, J Knol, J Tuynman, J Cucek, J Rojc and MR Rodríguez-Luna have no disclosures to make.