Prof. Dr.-Ing. habil. Jens Haueisen
Director of the BMTI Institute and head of Biomedical Engineering Group
Prof. Dr.-Ing. habil. Jens Haueisen
phone: +49 3677 69 2861
In biomedical engineering, we often encounter inverse problems where we want to infer the internal properties of the body from external measurement data. Examples from the Institute of Biomedical Engineering include (i) the imaging of pathological tissue changes in the human brain using quantitative magnetic resonance imaging, or (ii) the early detection of metabolic changes using fluorescence lifetime imaging ophthalmoscopy (FLIO) for the timely initiation of therapy to prevent blindness. The classical solution methods require the problems to be modelled as simply as possible. In most cases, the origin of the signals from the sources (forward problem) is already understood in much more detail than we can take into account with classical solution methods. Deep learning can implicitly derive complicated, non-linear inverse relationships from the training data. This leads to a paradigm shift because all knowledge can - and should - be incorporated into the model. Taking into account the smallest and most complex effects when evaluating biomedical data enables earlier and more accurate diagnoses and makes new things visible.
The central phenomenon of MRI imaging is magnetic resonance frequency precession. This frequency is shifted by a few millionths in biological tissue. The current state of the art is to attribute the frequency shift exclusively to magnetic susceptibility. Although other causes are known, it has not yet been possible to integrate them into the numerical methods. The aim of this project is to model additional biophysical mechanisms and thus enable the imaging of microstructure-related changes. The results will be used in the study and diagnosis of multiple sclerosis and other neurodegenerative diseases.
Serious eye diseases that can lead to blindness often begin with metabolic changes long before morphological changes in the back of the eye can be detected by ophthalmologists. Metabolic changes can be treated with appropriate medication. This can prevent permanent morphological damage to the retina and ultimately blindness. Early detection of incipient disease is therefore extremely important. Fluorescence Lifetime Imaging Ophthalmoscopy (FLIO) can be used to detect and monitor metabolic changes in the retina at an early stage. FLIO is not a direct imaging modality; the data must be processed to extract the fluorescence lifetime or features based on it to diagnose disease. This inverse problem typically involves a model-based process (tri-exponential model) in which iterative optimisation methods are used to estimate fluorescence lifetimes and other features. The accuracy of these estimation methods is highly dependent on the parameters used. In addition, these methods are computationally intensive and therefore time consuming. This complexity, together with the lack of standardised methods and the resulting error of fluorescence lifetimes, limits the further dissemination of FLIO, especially in the clinical setting.
The aim of this project is the AI-based estimation of fluorescence lifetimes from FLIO data. By using deep learning techniques, artefacts in the fluorescence lifetimes of the ocular fundus will be minimised, the fluorescence of the crystalline lens will be suppressed and the time required for this will be greatly reduced. A further aim is to reduce the number of fluorescence photons required for robust fluorescence lifetime estimation in order to reduce patient burden (particularly measurement time) and increase clinical acceptability.