Artificial intelligence for the advancement of multi-modality image registration.

Medical image registration methods tend to find a geometric transformation that would bring two images into spatial correspondence. Such geometrically aligned images provide information important for medical diagnosis or treatment. A traditional intensity-based approach does it by optimizing transformation parameters and maximizing image similarity. One of the main difficulties is the low efficiency and reliability of measuring image similarity for images of different modality (e.g. CT to MRI). In the past, we have proposed a point similarity approach that solves numerous problems, but such as all the other similarity measures it does still not optimally use all the knowledge that can be extracted from the data. We tend to bridge this limitation using modern approaches from the field of artificial intelligence, among which may be neural networks and reinforcement learning.

Keywords: Similarity measurement, Image registration, optimization.
Mentor: doc.dr. Peter Rogelj

EEG analysis for neurology and brain-computer interaction

EEG enables observation of brain activity but the useful information is hidden in a high complexity of signals. There are numerous questions related to neurology and human behaviour we could help answering by extracting useful information from EEG signals. In addition to this, we tend to develop novel brain-computer interfaces (BCI). The basic concepts are in both cases similar.

The topics for the final theses include:
– analysis of EEG data for neurological research tasks,
– development of novel approaches for analysis of brain connectivity,
– development of human-computer interfaces.

Keywords: EEG, signal processing, BCI.
Mentor: doc.dr. Peter Rogelj