Biomedical Machine Learning

This research line aims to develop generic machine learning approaches for automated image analysis, and to integrate these in clinical research as well as deploy them in clinical practice.

Important themes are the application of machine learning methods in radiation therapy, for image segmentation and registration, and dose prediction; in MR image acquisition and reconstruction; and in predictive analysis for disease and treatment outcome. Together with the Department of Radiotherapy we explore the area of adaptive radiation therapy (photon as well as proton, CT as well as MRI), which is an interesting setting for its demand for real-time solutions that are robust to real-life variations in patients. We explore deep learning methods for segmentation of target areas and organs-at-risk, for regression problems such as image registration, and also link the two. Together with the Gorter Center and in collaboration with Philips we develop deep learning methods for accelerated MRI acquisition and reconstruction. Prediction of patient and disease outcome is of growing interest.

Image registration  is an important technique in the field of medical image processing. It refers to the process of spatially aligning datasets, possibly from different imaging modalities (e.g. MRI and CT), different time points (e.g., follow-up scans or dynamic studies), and even different subjects. In many applications, this involves the estimation of nonrigid deformations. Examples are the modeling of patterns of development or degeneration (e.g. local growth or atrophy), estimating soft tissue motion and deformation, and assessing or compensating for between-subject anatomical differences. Part of our research focuses on deep learning solutions towards robust, real-time application, for example in a radiation therapy setting.

 

 

 

 

 

Current projects:

  • Open source image registration: the elastix toolbox
  • Development of Machine Learning Techniques in Vestibular Schwannoma Treatment: Quantification, Characterization, and Risk Stratification
  • Human-centric AI for contouring in head-and-neck cancer
  • Accelerated MRI acquisition and reconstruction
  • Machine Learning on mixed data from the clinic and from wearable sensors
  • New deep learning techniques for effective and robust radiotherapy planning
  • NextMRI: Truly portable MRI for extremity and brain imaging anywhere & everywhere
  • AI4AI: Artificial Intelligence for Accessible Imaging

Previous projects:

  • ADAPTNOW: High-Precision Cancer Treatment by Online Adaptive Proton Therapy
  • Esophageal Gross Tumor Volume Segmentation using Deep Learning
  • Registration visualization
  • Medical Image Registration – Linking Algorithm and User
  • Fast Image Registration for Time-critical Medical Applications
  • Brain MRI Image Analysis for an Ageing Population
  • Robust diffusion-weighted MRI for non-invasive monitoring of proton beam treatment 

Another important research focus is to ease the adoption of non-rigid image registration techniques in research and the clinic, by means of advanced interaction techniques, improved visualization, and high quality open access software implementation. The latter is realized through our image registration toolkit elastix.

Our Team members

  • Marius Staring, Team lead
  • Efe Ilicak, Senior researcher
  • Niels Dekker, Scientific Software Engineer
  • Laurens Beljaards, PhD student
  • Yunjie Chen, PhD student
  • Ruochen Gao, PhD student
  • Navid Jabarimani, PhD student
  • Donghang Lyu, PhD student
  • Prerak Mody, PhD student
  • Chinmay Rao, PhD student
  • Viktor van der Valk, PhD student
  • Jenia Makarevich, PhD student