Interactive virtual 3D models of renal cancer patient anatomies alter partial nephrectomy surgical planning decisions and increase surgeon confidence compared to volume-rendered images

The evolution and uptake of laparoscopic and robotic-assisted techniques have led to an increase in the use of minimally invasive partial nephrectomy (PN), also known as nephron-sparing surgery, for the treatment of the small renal mass [1, 2, 3]. A key driver behind this trend is the accumulating evidence that post-operative healthy kidney volume is positively correlated with improved kidney function as measured by the estimated glomerular filtration rate (eGFR) [4, 5]. Surgical planning for minimally invasive PN is complex, with numerous patient and tumour characteristics having to be accounted for, especially the relationship between the tumour and renal hilar anatomy. Historically, the appreciation of these anatomical factors has been through the examination of coronal, sagittal, and axial slices of computed tomography (CT) and magnetic resonance imaging (MRI) datasets. Recently, the use of volume-rendered images (VRI) for surgery planning has become more common [6], but VRI has significant limitations. VRI can only reveal structural information when tissue contrast is high, VRI can only be applied to a single scan at a time, and VRI supplies no semantic knowledge. Therefore, quantitative analysis such as distance/area/volume measurements cannot be undertaken and typically the relationship of the tumour to collecting or venous systems cannot be appreciated when using VRI.

To overcome the general limitations of VRI in a soft tissue oncology setting, dedicated software packages have been developed to classify medical scan voxels into their anatomical components in a process known as image segmentation [7, 8, 9]. Once segmented, stereolithography files are generated which can be used to visualise the anatomy and have the components 3D printed. It has been shown that such 3D printed models influence surgical decision-making and patient education [10, 11]. However, the relevance of a physical model to plan for a minimally invasive surgical approach is debatable, and the financial and administrative costs of obtaining accurate 3D printed models for routine surgery planning have been speculated to be holding back 3D printed models from breaking into regular clinical usage [12].

As a necessary precursor to 3D printed models, computational or virtual 3D models could be used by the urologist to assist with clinical decision-making. Virtual 3D models provide many of the advantages of their physical 3D printed counterpart without the challenge of the printing process. They can be easily viewed on standard digital devices such as laptops or smartphones, and they can be simultaneously viewed from anywhere in the world which could help with collaborative surgery planning between centres. Furthermore, the ability of the viewer to interact with the virtual 3D models, e.g. selecting which anatomical components are visible or setting preferred viewpoints, intuitively should lead to efficiencies in learning the important anatomical features for the given clinical application. For example, pioneering studies have already shown that surgeons benefit from virtual 3D models in the theatre, particularly with regard to improved appreciation of hilar vascular anatomy and noteworthy attempts to minimise ischaemic period and volume through enhanced pre-operative planning and simulation [13, 14]. However, in addition to the available routine imaging (CT, MRI, VRI), it has not been shown that virtual 3D models would influence the surgical decision-making process or alter surgeon confidence in their decisions.

The objective of the current study is to determine whether the interactive visualisation of patient-specific virtual 3D models of the renal anatomy influences the pre-operative decision-making process of urological surgeons for complex renal cancer operations. A secondary objective is to determine whether or not surgeon confidence levels can be increased by the addition of virtual 3D models, which could have a positive impact on surgeon stress levels [15].

Pre-operative planning for renal cancer surgery is critical for achieving the best patient outcome. While mindful of key patient co-morbidities, the surgical plan hinges on surgeon awareness of the often complex spatial relationship amongst the various anatomical components. Tumour localisation, nearness to the collecting system, and vascular invasion are all important factors that are not necessarily easy to correctly grasp from the constrained 2D visualisation of CT or MRI data alone. This study aimed to demonstrate a possible improvement in the tools available to the urologist for surgery planning and to investigate how the proposed technology compares to the status quo.

This study is not the first time that virtual 3D anatomical models have been used to aid surgery planning. For example, multiple research groups have previously introduced such models in the Operating Room (OR) via the da Vinci™ console using the TilePro™ function (Intuitive Surgical, Sunnyvale, USA) [13, 17]. Although small-scale proof-of-concept studies, they showed that the use of virtual 3D models in the OR did not distract from the procedure and provided useful additional information for surgery guidance purposes. Similar to 3D printed models [18], the results of this study provide further evidence that virtual 3D models when used as an adjunct to surgery planning can improve surgeon understanding of the patient anatomy, as indicated by the significant increase in the surgeon’s opinion of their own anatomical awareness. Arguably the greatest benefit of virtual 3D models is for atypical cases, however, such as high number of renal arteries or horseshoe kidneys [19]. The horseshoe case featured in this study (case D) was the only case to obtain more than a 1-point increase in median surgeon confidence in their segmental clamping plan. It is worth noting that due to a lack of resources or technology not all surgeons have access to volume-rendered images for surgery planning assistance. It could be reasonably assumed that the extra benefit of virtual 3D models from a planning perspective would be even greater than demonstrated in this study when compared to the original multiphase CT scans alone.

This study also investigated the impact of virtual 3D models on the key clinical decision of whether to use an open, laparoscopic, or robotic surgical approach to partial nephrectomy. A change in surgical approach was made in almost 1-in-5 cases (19%) with the only difference between study arms being the form of the imaging information provided. Notably, two-thirds of those approach changes were from an open to a minimally invasive technique, suggesting the potential for treating more patients with nephron-sparing surgery under the interventional imaging protocol. This result provides a clear pathway to impacting patient outcome by using enhanced surgery planning tools.

The clinical utility of virtual 3D models was consistently judged to be significantly greater than that of volume-rendered images (VRI) across the three clinical stages considered: multidisciplinary team meeting, theatre planning, and intra-operative for assisting navigation. Despite the preference for virtual 3D models, the use of such models for surgery planning has still not been translated into common practice. This is likely due to the requirement of significant labour and expertise in manually constructing the virtual 3D models from the original medical scans, as well as the financial cost [20]. This bottleneck of obtaining the pre-operative anatomical segmentation underlies the main challenge to many forms of quantitative surgical assistance, from 3D printing of orthopaedic implant guides [21] to forming the baseline, pre-operative model for augmented reality intra-operative surgical navigation systems [22]. Advances in machine learning techniques, typically using convolutional neural networks (CNNs), have made significant inroads into overcoming this barrier [23, 24]. Our group have recently applied similar CNN methods for fully automated kidney segmentation from contrast-enhanced CT and achieved state-of-the-art Dice scores of over 95% for both the left and right kidneys [16]. Continued improvements in computer vision image segmentation algorithms should enable similarly accurate models to be constructed for all abdominal organs within clinically relevant timeframes, which is already obtained using slower computational methods [25]. Extensions of this work to more challenging aspects such as abnormal tissue classification [26], vessel segmentation [27, 28], and methods of efficiently incorporating user interaction to improve model generation are underway across several research groups [29]. Moreover, scalable virtual 3D model generation could lead to the adoption of improved methods of not just surgery planning but also patient-specific pre-operative training [30].

This study has its limitations. Firstly, it featured analysis from 25 respondents using a qualitative questionnaire. The number of respondents is in line with or larger than other similar studies [10, 13], but extrapolation to the general urologist population should be treated with caution. Secondly, the study’s 54% response rate may also result in selection bias with respondents perhaps being those most amenable and interested in the use of virtual 3D imaging. Thirdly, study data analysis was mostly restricted to simple frequency and median statistics due to the use of qualitative Likert items [31], but statistical power was achieved by considering all 125 surgeon dataset combinations for testing overarching hypotheses not specific to any one patient. Furthermore, the study design of comparing the intervention to control imaging renders analysis of inter-rater agreement to be nonmeaningful. Specifically, the use of virtual 3D models made surgeons generally believe that they had a better spatial awareness of the patient body which resulted in a majority of high Likert scores causing an unbalanced results dataset which is unsuitable to Fleiss Kappa statistical analysis. Fourthly, the virtual 3D models are not yet perfect representations of the patient anatomy. For example, in Fig. 1b (right) the combination of finite spatial resolution of the image and partial volume effects has resulted in two separate, traversing arteries to appear fused. More algorithmic research is required to best handle edge cases such as vessels of the class being in extreme proximity. Finally, this study did not investigate clinical outcome. While our results clearly indicate that surgeons are absorbing anatomical information via the virtual 3D models in a different manner to VRI leading to a distinct change in surgical plan in a fifth of cases, it does not prove that the approach taken with the use of the virtual 3D models is necessarily a better choice in terms of patient outcome. A prospective randomised controlled trial (RCT) examining primary outcomes such as length of stay, complication incidents, blood loss, and OR time will be the ultimate arbiter for healthcare outcomes impact.

This study demonstrates that the use of interactive virtual 3D models as an adjunct to renal cancer surgery planning influences clinical decisions which can result in changes to surgical approach decisions, as compared to planning with the status quo imaging. Compared to volume-rendered images, virtual 3D models were deemed by urologists to present the patient anatomy in a clearer manner, to increase surgeon confidence in their surgical plan, and to have a higher potential for clinical benefit at MDT, theatre planning, and intra-operative stages.