Master's Thesis presentation, Sofie Allgöwer and Sofia Ljungdahl

The standard treatment of liver cancer tumors in acute cases is surgical resection performed by either open surgery, where the abdomen is cut open, or laparoscopic surgery, a minimally invasive procedure. To navigate during laparoscopic liver surgery today, the surgeon switches between looking at the laparoscopic camera view and preoperational images and creates a mental image of where the tumor is located in the liver. To support the surgeon during navigation, the start-up Navari Surgical AB is developing an augmented reality (AR) solution offering intraoperative guidance during minimally invasive procedures. The solution provides an AR projection of the tumor in the camera view. This thesis work is executed in conjunction with Navari and aims to contribute with a guideline for how tumor segmentation in computer tomography (CT) scans could be performed to be able to create the AR projection of the tumor.

To perform tumor segmentation in CT scans, both classical algorithms such as thresholding and active contour models, as well as a machine-learning solution, have been investigated. The best results for the algorithms achieved a dice score of 0.4 and a recall value of 0.6. For the machine learning approach, the U-Net architecture was utilized and showed promising results with a dice score of 0.77 and a recall value of 0.8.

From the research and investigation of the different methods for liver tumor segmentation, the results indicate that machine learning is the most promising approach. However, further research is needed to achieve a satisfactory result, with many potential areas of improvement.

Opponents: Bahareh Rezaei and Tasnim Ahmed