Navisurg: Decision Support and Navigation Assistance for Safer Image-Guided Surgery In

NAVISurg: Decision support and navigation assistance for safer image-guided surgery in pancreatic interventions

P. Sánchez-González, I. Oropesa, B. Rodriguez-Vila, J.A. Sánchez-Margallo, E. Cueto, J.L. Peris,
J. Mayol, F.M. Sánchez-Margallo, E.J. Gómez

Purpose

Pancreatic cancer has one of the highest fatality rates of all cancers, and is the fourth-highest cancer killer worldwide. Moreover, the incidence of pancreatic cancer has increased constantly since the 50’s. It is estimated that each year about 233,000 new cases of pancreatic cancer are diagnosed worldwide, 60% of them in Europe, North America and Australia.

Under the tag of pancreatic cancer, several different types can be differentiated depending on their origin and location, and the surgical approach may differ in each case. A coarse categorization depending on the lesion origin distinguishes between exocrine pancreas cancer (where the most common form of pancreatic cancer, ductal adenocarcinoma, lies in), pancreatic cystic neoplasms and pancreatic neuroendocrine tumors. Additionally, the location can be tagged as head, body or tail.

The pancreatic cancer staging and characterization is extremely complex, and often the preoperative procedures do not offer a conclusive diagnostic. Due to the limitations in conventional medical imaging and the restrictions imposed by both the anatomy and the surgical approach, there is a need for methods that support intraoperative imaging and decision making in order to improve their accurate anatomical localization, the characterization of their nature and the optimal surgical approach.

The main goal of NAVISurg is to develop an intelligent surgical navigator for pancreatic interventions. The navigator is based on the Image and Video Guided Surgery (IVGS) paradigm. The main pillar of the paradigm contemplates the exploitation of the information from the endoscope (ever-present in the laparoscopic surgical scene) to update the preoperative information of the patient to the reality in the OR. The navigator will additionally incorporate a clinical decision support system (CDSS) which, based on patient record, preoperative studies and intraoperative imaging, will provide counsel on the best therapeutic approach to follow.

Methods

Research areas contemplated in the definition of the NAVISurg navigator are focused on (1) pre-and intra-operative pancreas segmentation and reconstruction; (2) implementation biomechanical model of the pancreas; (3) intraoperative update of the biomechanical model and (4) decision support system for diagnosis, treatment and prognosis.

Preoperative segmentation is performed at two levels: (1) organ surface and (2) pancreatic duct tree. It is based either on MDCT or MRCP studies. Segmentation, performed using an Active Appearance Model (AAM) based on 19 pancreas CT images and their related segmentations, and reconstruction of the 3D model constitutes the basis for the patient-specific biomechanical model.

Accuracy and resolution of the preoperative 3D reconstruction needed for the biomechanical modeling are specially considered. Ideally, everything is modelled with continuum-based finite elements, with each element being assigned with the correct mechanical properties. Because model-output should be available online during the operation, solutions should be calculated in real- or near-real-time. Due to computational restrictions, justifiable simplifications are made, including model reduction approaches.

Intraoperative update is based both on LUS and endoscopic video. LUS imaging is used to obtain a volumetric update of the intraoperative pancreas state, and can be both used for initial intraoperative registration or during key moments of the surgery. LUS segmentation of pancreas surface and internal structures (tumors) is based on edge enhancement and level sets, while freehand 3D LUS reconstruction is based on the tracking data from an Aurora system located at the tool tip. On the other hand, endoscopic video (an always present source of imaging during the intervention) is employed to provide a superficial update of the biomechanical model. The underlying hypothesis is that changes in the model surface can be propagated to its volume in this way. Reconstruction of the surgical scene is addressed by means of shape from shading and shape form motion techniques.

In order to build the CDSS, a decision tree modeling the current knowledge on pancreatic cancer and the diagnostic and therapeutic options in use for the different pancreatic tumors is being built. The decision tree will be used to train classifiers for diagnostic and treatment of the lesions. In order to incorporate data from the patient image studies (preoperative and intraoperative), content-based multimedia retrieval techniques are being employed.

Results

Preliminary advances in the different research areas of NAVISurg include:

Preoperative segmentation and reconstruction of a porcine model of the pancreas, based on level sets using ITK-Snap (Figure 1).

Figure 1. Ex-vivo porcine pancreatic model and its 3D reconstruction

Human pancreas biomechanical model. A model has been created for the pancreas and a portion of duodenum for appropriate boundary conditions. Pancreas has been modelled as hyperelastic and neo-hookean (Figure 2).

Figure 2. Preliminary results of the biomechanical models for human pancreas and duodenum

Intraoperative segmentation and reconstruction of the pancreas based on LUS from an ex-vivo experiment.
Shape form motion and shape from shading techniques, able to achieve a surface reconstruction of the surgical scene and the 3D localization of the endoscope during the procedure.
A decision tree modelling pancreatic lesions and interventions based on an extensive up-to-date review of the clinical state of the art.

Conclusion

The development and validation of the NAVISurg navigator will contribute to prove that preoperative and intraoperative medical images, along with video information, are useful both for pancreatic cancer staging, tumor characterization and effective navigation during tumor removal. Moreover, with NAVISurg platform, decision-making support could be made available to less experienced surgeons in these specific surgical procedures. In addition, it would assist in assessing tumor resectability and in providing image-guidance for the procedure. It is expected to result in more effective, less aggressive, more efficient and safer pancreatic procedures.

Acknowledgments

This work has been partially funded by NAVISurg project under the CIBER-BBN intramural research program.