Lisanne van Dijk

UMC Groningen

In short, Lisanne van Dijk's research entails the following: 

Deep learning to Enhance Therapeutic Outcomes and toXicity for LUNG cancer patients (DETOX-LUNG)   

'Radiotherapy is an important treatment for lung cancer patients, yet they are often faced with severe toxicities due to radiation damage. These toxicities do not only have a big impact on the quality-of-life of patients, but can also have a negative impact on their prognosis. Adequate prediction of treatment response is needed to prevent severe toxicities and (related) fatalities by personalized treatment optimization.

This project aims to improve the lung cancer radiotherapy outcome prediction with Artificial Intelligence. Artificial Intelligence has the capacity to deal with three-dimensional radiation dose admitted to the patient for the prediction of toxicities, instead of the current models than can only deal with an average radiation dose value to a single organ. This can be combines with the patient-specific characteristics from medical images and clinical data.' 

Short summary of progress/results March 2025

We are developing advanced artificial intelligence (AI) models to better predict side effects and survival for lung cancer patients treated with radiotherapy. Our recently developed deep learning model uses full 3D scans and treatment data to predict radiation-induced lung damage (pneumonitis) more accurately than current conventional models, which rely on simpler, less detailed information (AUC: 0.79 vs. 0.72).[LD1] This can help doctors make safer, more personalized treatment decisions. We’re currently working on adding PET scan data to further improve predictions by adding physiological information. Additionally, we are now ready to start developing the deep learning to predict patient survival by combining imaging with clinical data. These promising results of the pneumonitis AI models are ready for publication and represent an important step towards more tailored and effective lung cancer care.

Short summary of progress / results March 2024

For training the Deep Learning model, a large amount of patient data is needed. This collection and verification of the data are essential for all subsequent steps of the research project. Over the past months, data collection and processing have taken place using clinical and toxicity data from RedCap, as well as DICOM files containing CT scans, radiation dose distribution, and segmentation data of the tumor and critical organs. In total, a dataset of 1107 patients has been obtained for predicting radiation pneumonitis (WP1) and overall survival (WP2). Currently, we are in the process of developing the reference, radiomics, and deep learning models. The collection of PET images (WP3) has been started in parallel. Noteworthy, we have some delay, due to complications in the hiring process.