Michiel van den Brekel

NKI

In short, Michiel van den Brekels' research entails the following: 

Genomics-assisted radiomic machine learning-based models to enable patient outcome prediction for personalized treatment in head and neck squamous cell cancer

'This project will be carried out in the Netherlands Cancer Institute with generous funding by the Hanarth Fonds. Previously we have developed models, using genetic as well as radiomic signatures, to predict outcome of patients treated for head and neck squamous cell carcinomas. It turns out that specific, pathway related genetic signatures can predict response to for example radiotherapy or chemotherapy. This knowledge will eventually lead to more personalized treatment. The radiological features, are so far less well explained by biological processes. The primary goal of this project will be to build multimodal models (combining multiple imaging modalities and genetic data), and signatures for use in developing outcome prediction models in advanced head and neck squamous cell carcinoma. One of the main aspects of the imaging biomarker search will be investigating the link between molecular tumour biology characteristics, as captured by genomic data, and the tumour morphological phenotype, as quantified by the imaging data.'

Short summary of progress / results February 2024

A manuscript titled "Overcoming data scarcity in radiomics/radiogenomics using synthetic radiomic features" was submitted to the journal 'Computers in Biology and Medicine', the first significant step. Following this, the abstract was accepted for presentation at the European Congress of Radiology (ECR) 2024. In a related development, external validation of a previously published radiogenomics model was completed, yielding valuable results. This led to the preparation of a second manuscript, "Multicenter validation of an MRI-based radiogenomic model predicting human papillomavirus status in oropharyngeal cancer". The team also conducted an analysis on the impact of synthetic data generation models on a multicenter oropharyngeal cancer cohort, revealing promising preliminary results. A new study design was proposed, and a master's student was hired to benchmark feature selection methods and develop a novel algorithm for head and neck cancer radiomics/radiogenomics. The IRB approval was obtained to start a retrospective study and the delineation of MRI scans at the Netherlands Cancer Institute (NKI) for this study. 

Short summary of progress / results February 2023

In this project, patient data from NKI-AVL and AUMC have been selected and are currently studied. The selection techniques and algorithms to synthetise radiomic data, based on a large database of patients with cancer, all delineated on MRI and genotyped, is very promising, and probably can also be used in areas with more data scarcety. This can help to improve feature selection in similar studies.