Medical Informatics, Artificial Intelligence and Deep Learning
Our institute leverages state-of-the-art methods inlcuding the application of deep learning modelling strategies in medical imaging and the use of artificial intelligence to facilitate handling of large-scale data.
Deep learning and medical imaging
In the context of experimental brain images, the registration of histological images to an atlas is still challenging in terms of accuracy, universality and time efficiency. The first issue is that brain histological datasets often suffer from artefacts, such as enlarged ventricles (holes), missing tissue, folding, air bubbles, uneven staining, tears or slice-independent distortions. Second, due to brain tissue elasticity, the sections are easily deformed during slice preparation procedures. Consequently, when computing this warp field, the elasticity of different regions in the brain and its mechanical properties must be taken into account. Third, because of the warped nature of the histological brain images, there is no ground truth data for the boundaries of the anatomical regions.
We investigate deep learning methods to register histological mouse brain slices, using a standardised annotated mouse brain atlas. The accompanied software provides a deep learning-based registration method, which can be evaluated by means of a validation measure in comparison with ground truth data. Additionally, it can be used to explore ways to simulate ground truth data, which can be used to validate the registration methods of the software.