In this pilot study, 3D VR-enhanced abdominal CT and MRI were able to display the anatomy in an understandable way, both for junior and experienced medical professionals. The VR experience was received positively by the tested study population. Experienced surgeons and radiologists were able to assess the vascular involvement and, ultimately, the resectability of PDAC in the majority of cases presented. Direct comparison between 3D VR enhanced and standard 2D PACS viewing of CT imaging displaying PDAC showed significantly higher accuracy in the VR group.

The first part of the study aimed at assessing the understandability of 3D VR enhanced cross-sectional abdominal imaging. There are only a few studies that directly assess the comprehensibility of VR-enhanced abdominal CT or MRI anatomy [15, 16]. Previously, a VR-enhanced CT scan was reported in a feasibility study, however, the software automated segmentation and PACS import was not available. For the implementation of VR as a clinical adjunct, the accuracy and comprehensibility of VR-enhanced visualisation will need to be further scrutinised to guarantee patient safety. This study aims to lay a cornerstone for this purpose, as most participants, including junior doctors could reliably identify the displayed anatomy. In the past, this VR software has been evaluated as a tool to teach anatomy, indicating that anatomical learning was perceived as more efficient and engaging when compared to the standard anatomy learning with models or books [17]. A recently published meta-analysis including a total of 15 randomised trials that evaluated the efficacy of VR-supported anatomy teaching, concluded that post-VR intervention anatomical test scores were significantly higher when compared with other teaching methods [9]. In these studies, anatomical VR models were used, but not PACS imported imaging datasets. Overall, these results align with previously published results from our group, showing that VR MRCP led to faster and more accurate anatomical understanding when compared to printed MRCP scans [5].

Although the participant’s feedback was largely positive, and more participants provided positive than negative feedback, a few criticisms must be addressed. The main criticism of the VR experience was a perceived lack of image resolution. The used VR software imports isovoxels from the original dataset in a 1:1 fashion, thus the resolution is limited not by the VR software, but by the original dataset. In VR, the model can be zoomed in and magnified to an extent where the VR model becomes larger than the scanned person in real life. By comparison, most screens used are relatively small, leading to the viewers perception of sufficient image resolution as compared to the magnification possible in the VR system. In the future, artificial intelligence might be introduced to improve appearances and resolution of imaging, however, this may come with new difficulties and problems as the original dataset could be altered [18]. Another comment issued by the participants was the weight of the headset. Although a recognised problem, recent advances in hardware development will likely render this issue obsolete and should not be seen as a deterrent for future use of VR in a clinical setting [19].

In the second part of the study, this VR software was evaluated as a tool to assess the presence or absence and extent of vascular contact, and ultimately the resectability in PDAC. In general, inter-observer agreement is known to be low in PDAC imaging and has been reported to be as low as 7.2–30% [10, 11]. This is also true for experienced radiologists, who show only slightly improved interobserver agreement compared to their less experienced counterparts [11]. Given these numbers, our reported results in the 3D VR group show a relatively high interobserver agreement. This is also reflected in the Fleiss κ range of 0.4–0.7 in our cohort, compared to the κ range of 0.282–0.555 reported by Giannone et al. [10] Of note, the interobserver agreement in the standard 2D PACS group was within this previously reported range. Furthermore, median time needed by the participants to reach a conclusion was only 185 s (VR CT) and 116 s (VR MRI), representing a fast assessment. It is important to note that the study participants were free to use as much time to assess the scans as they liked, and interacted significantly shorter with the standard PACS imaging. The shorter viewing time in the 2D PACS viewing group may be attributed to the individuals’ familiarity with standard 2D assessment of scans. Participants are accustomed to this format which could lead to faster assessments. In contrast, 3D VR viewing invites the viewer to study the scans in more detail, using various tools at hand to freely interact with the scan from multiple angles. Increased exposure time could also be a part of the explanation for improved results in the 3D VR group, as participants take more time to study the scans before coming to a decision. To further assess the value of VR enhanced imaging to evaluate resectability of PDAC, a larger randomised prospective trial is needed, and correlation with perioperative as well as long-term clinical outcomes could shed light on the clinical value of this technology. Confirmation of these findings in a larger trial could support the transition of standard 2D PACS viewing to routine use of VR enhancement technology clinical practice.

As previously reported, interobserver agreement increases especially in cases of borderline resectability [20]. One of the main reasons contributing to this is that resectability represents a continuum rather than clear groups and might also be judged differently by individual surgeons. This issue has been recognised and recently a debate to redefine the terms ‘resectable’, ‘borderline resectable’ and ‘locally advanced’ has been proposed [10]. In our study cohort, comparison of interobserver agreement between the different resectability categories would require a larger study. Analysis of the variation between official MDT report and assessment of vascular infiltration and resectability status in VR reveals roughly three categories of errors. In the first category, imaging quality was insufficient or misleading, as in the case with the previous right hemicolectomy and the metal artefact. This scan was left in the final analysis to omit an imaging selection bias. The second category includes scans which allow for varying interpretations, such as the scan with the abutment of the SMA slightly below 180°. This scan was identified by one participant as R, but according to the guidelines and the MDT decision was seen as BR. The third category includes scans where the participant was not able to identify the tumour as such. Recently, sensitivity for CT and MRI to detect PDAC have been reported to be 75% and 70%, respectively [21]. Nonetheless, participants were able to identify PDAC and its resectability status overall rapidly and reliably, even surpassing the numbers reported in the literature. This is also taking into consideration that the participants were not accustomed to VR image viewing and have not had the chance to complete their learning curve when handling the VR system, although its use is largely intuitive.

A further added difficulty of resectability assessment based on imaging is the introduction of neoadjuvant chemotherapy in BR PDAC, as cross-sectional imaging appears to overestimate residual cancer and underestimate resectability [22]. Whether VR technology can improve assessment post-neoadjuvant chemotherapy could be a rewarding question for future studies in this context.

An interesting result from our study was that the pancreas was not identified on VR CT by all participants, which however did not represent an issue in the second study step. This is likely explained by the fact that also relatively inexperienced colleagues participated in the first study step, whereas only experienced surgeons and radiologists as well as surgeons who underwent subspecialist HPB training were included. Therefore, this result might reflect on the experience of reading cross-sectional imaging data rather than represent a weakness of the technology per se. In general, a lack of recognition of the tumour does not necessarily reflect a shortcoming of the technology but might be linked to the viewer’s experience and understanding. However, no correlation in the second study step was found between work experience and number of correct answers. Based on this finding, it can be assumed that clinicans can potentially use technology regardless of their experience. Of note, participants in this study preferred the use of VR CT over MRI, which could be explained by the fact CT imaging is widespread in daily clinical practice and surgeons might feel more confident interpreting these scans. However, the number of study participants is too low to draw a meaningful conclusion in this regard. Overall, however, there is controversy surrounding the value of MRI and CT for the assessment of PDAC, and it appears that MRI has a slight advantage over CT in terms of tumour detection rate, but CT appears to be superior for the assessment of vascular infiltration [23]. In this study, we have not included positron emission tomography (PET), and adding such scans to a future study could potentially increase the value of the VR system as tumours could be more easily identified.

The main limitations of this study are the relatively small number of participants in the second study step. Although VR enhanced CT imaging lead to better understanding of resectability of PDAC, a randomised prospective study, with a larger number of participants that directly compares VR to standard imaging and incorporates intraoperative outcomes, especially in BR cases, could offer further information on the clinical use of this software. A further future use of this technology could lie in patient education with the VR visualisation to demonstrate the imaging data to the patients, allowing for a better understanding of their disease and its treatment. Furthermore, implementing this technology in MDT could potentially yield effects such as enhanced collaboration among team members of different specialties, a more comprehensive analysis of medical imaging data and improved communication in case discussion. It is possible that the definitions of resectability will change in the future, and for this reason the implementation of a novel VR technology should not solely be based on the number of interobserver agreement, but rather also be based on clinical utility and user experience. In the future, connecting VR-enhanced visualisation data with AI algorithms that allow for automated assessment of vascular tumour contact might even revolutionise resectability assessment and preoperative planning [24, 25].

In conclusion, VR-enhanced CT and MRI visualisation is a promising tool to display abdominal anatomy that is easy for clinicians of varying levels of experience to interpret. Furthermore, there is the potential for VR-enhanced CT and MRI to be implemented as a tool to aid in future surgical planning and oncological treatment of PDAC.