After cardiovascular diseases, cancer is the second leading cause of death, with the liver being frequently affected by primary and metastatic tumors. Due to the complex and highly variable liver anatomy, liver surgery requires careful patient-specific preoperative planning based on CT or MRI data. For this purpose, 3D models of the liver, vessels, tumors, and bile ducts are generated. However, resection planning is still often performed manually or semi-automatically, making the process time-consuming and user-dependent.
The LIZARD project aims to develop deep learning–based methods for segmentation, surface model generation, and deformation to support patient-specific resection planning in liver surgery. A key component is the creation of a training database that also enables classification of liver anatomy. In addition, clinical data are integrated to improve the methods and to account for different tumor entities as well as multiple lesions.
The project focuses on developing interpretable deep learning strategies that enable liver surgeons, radiologists, and clinical researchers to reliably predict liver resection volumes. The models will be clinically evaluated and refined using expert feedback. In the long term, the project seeks to provide a fast and easily accessible solution that does not require expensive hardware or highly specialized personnel. This will support better patient assessment and referral to specialized centers, ultimately enabling more patients to receive potentially curative treatment.
