Cornelis van den Berg

UMC Utrecht

In short, Cornelis van den Berg's research entails the following:

Improved residual disease detection after (chemo-)radiotherapy for locally advanced head and neck squamous cell carcinoma

For advanced head and neck cancer, the usual treatment involves preserving organs through (chemo)radiotherapy. However, there's a risk of about 20% of residual disease (detectable tumor) six months after treatment. Evaluating treatment response, whether clinically or with imaging, is challenging due to radiation effects mimicking tumor activity. This project aims to create an AI-based clinical decision support tool for radiologists and radiation oncologists. It will help assess tumor response and predict the risk of recurrent disease within the first two years post-treatment. The goal is to reliably distinguish between residual disease and post-treatment changes, improving the accuracy of assessing the need for salvage surgery and can individualize follow up schemes. This could enhance survival and quality of life while minimizing unnecessary procedures and reducing patient burden.

Short summary of progress/results April 2026

We are developing AI methods that can automatically spot unusual patterns in medical images without needing large amounts of labeled data. As a first step, we have shown that this approach for detection of primary tumors in brain and head-and-neck cancer scans. We are now extending this work to identify cancer recurrence in patients after treatment. To do this, we are exploring advanced AI techniques, such as modern self-learning models. This work has already been presented at several scientific conferences.

At the same time, we are collaborating with international partners to collect a large dataset of diffusion MRI scans. These scans provide information about tissue structure and may help predict what the chance is a tumor miht come back after treatment. Previous studies have shown mixed results, likely due to limited data. By combining data from multiple centers, we aim to perform a more robust analysis and provide clearer, more reliable conclusions.

We have demonstrated self supervised AI methods for tumor detection and recurrence localization in brain
cancer and H&N cancer.

Short summary of progress/results April 2025

A PhD student has joined a project aiming to develop an AI tool to assist clinicians in detecting tumor recurrence in head and neck cancer. The initial months focused on training in medical imaging, radiotherapy, and AI. Since October, the focus shifted to collecting and curating imaging and clinical data from UMC Utrecht. Data from 358 oropharyngeal and 318 laryngeal cancer patients were reviewed The first goal is to build an anomaly detection model using a variational auto-encoder trained on non-recurrent cases, to highlight abnormalities in recurrent patients.  Additionally, using this dataset the team is validating a previous study that include a small dataset suggesting that pre-treatment MRI (ADC values)could predict recurrence.