Kristian Overduin

Radboudumc

In short, Kristian Overduin's research entails the following:

AI-IMAGINE - Automated Intraoperative assessment of IMAGINg Endpoints for first-time right liver thermal ablation

Thermal ablation is a minimally invasive treatment in early-stage primary and low-volume secondary liver tumors. Compared to surgery, ablation is safer, cheaper and associated with faster recovery, however local recurrence rates tend to be higher, leading to unwanted re-treatments and limiting its use as primary treatment.

Prior studies have demonstrated quantitative 3D treatment margins derived from intraprocedural computed tomography (CT) images to accurately predict the risk of local recurrence after thermal ablation of liver tumors. However, traditional manual image analysis methods to derive the treatment margin are subjective and time-consuming, limiting its generalizability and intra-operative applicability.

The AI-IMAGINE project will investigate AI to develop an automated method for intra-operative evaluation of treatment margins and prediction of treatment success in thermal ablation treatment of liver tumors. This empowers clinicians to better assess treatment success during the actual treatment itself and, in future patients, allow immediate corrective treatment. Hereby, we expect to reduce local recurrence rates and avoid unnecessary re-treatments. Ultimately, we aim at offering patients a minimally invasive treatment that is as effective as surgery, but is safer, cheaper and has less patient impact.

The AI-IMAGINE project is an international collaboration between the Minimally Invasive Image-guided Interventions (MAGIC) and Diagnostic Image Analysis (DIAG) group at Radboudumc and several academic institutions (Leiden UMC, Amsterdam UMC and Medical University Innsbruck, Fraunhofer MEVIS).

Short summary of progress/results 2025

The first part of the AI-IMAGINE project consists of a multi-institutional reader study evaluating human performance in ablation margin quantitation and local recurrence prediction in thermal ablation of hepatocellular carcinoma and colorectal liver metastases by 4 independent readers. The analysis is ongoing and projected to be finalized in Q3 2025.

In the second part of the project, we have been simultaneously developing base models for AI-based automated tumor and ablation zone segmentation, firstly in colorectal liver metastases. Initial results show good performance for fully-automated ablation zone segmentation and moderate performance for tumor segmentation. Initial results will be presented at European Conference on Interventional Oncology 2025. Model improvement strategies and model development for hepatocellular carcinoma are currently being investigated.