Introduction
Pancreatic cancer is currently the fourth leading cause of cancer death in the United States and has only a 5-year survival rate of 10%. Pancreatoduodenectomy is the cornerstone of curative treatment for patients diagnosed with pancreatic head cancer. Only 20% of the patients are candidate for surgery since many patients present with distant metastases or locally advanced tumours with vascular involvement. Whether tumours are amendable for resection is based on potential vascular involvement with the surrounding vasculature. Assessment of vascular involvement, mainly based on multi-phase computed tomography (CT) scans, requires specific expertise and can be challenging. Computer-aided detection (CAD) and autostereoscopic three-dimensional (3D) patient models might improve accuracy in predicting vascular involvement and improve overall surgical planning. This graduation project aims to assess the added value of autostereoscopic three-dimensional patient models and computer-aided detection for decision support in pancreatic cancer care.

Technical methods
An integrated medical imaging workstation was developed based on the clinical needs of clinicians regarding key concepts of the preoperative planning of pancreatoduodenectomy. This workstation is a hardware and software combination that consists of three main components; 1) a medical imaging viewer showing CT scan with basic imaging functionalities, 2) annotations outlining the tumour and anatomical structures (to simulate segmentations generated by CAD algorithms) that are translated to 3D patient models displayed on an autostereoscopic display, and 3) CAD-derived metrics (degrees and length contact) regarding vascular involvement of the tumour.

Clinical methods
This integrated medical imaging workstation was evaluated in a multi-centre study including 13 expert hepatopancreatobiliary surgeons and one abdominal radiologist. All participants assessed pancreatic tumours in a simulated setting under 3 different test conditions; assessment using only the regular CT scan (CT-condition), assessment using CT and 3D patient models (3D-condition), assessment using CT, 3D patient models and CAD-derived metrics regarding vascular involvement (CAD-condition). A total of 6 patient cases were evaluated, of which 3 radiologically resectable cases (simple) and 3 radiologically borderline resectable cases (complex) with pancreatic tumours near to major vessels. Perceived fulfilment of clinical needs regarding the preoperative assessment, differences in surgical planning decisions compared to baseline, and confidence in clinical decision-making were evaluated.

Results
Clinicians experienced an improved ability to accurately detect pancreatic tumours and determine the degrees and length of tumour-vessel contact under the 3D- and CAD-condition compared to the CT-condition. Additionally, clinicians reported a higher perceived ability to identify, localize and understand anatomical relationships when supported by autostereoscopic 3D models. Lower degrees of tumour-vessel contact were reported under the CAD-condition compared to the CT-condition. Furthermore, clinicians had higher confidence in assessing the need for a vascular resection under the 3D-condition than the CT-condition.

Conclusion
CAD and 3D might improve the accuracy of pancreatic tumour detection and reduce the overestimation of degrees of vascular involvement on radiological imaging. The risk of over-trust in CAD mandates thorough evaluation of the accuracy and use of CAD in prospective studies.