Magnetic Resonance Images (MRI) are of high interest for surgical operations (e.g. biopsy, ablation, etc.) due to its non-invasiveness, high resolution and contrast; in particular, when needles and probes are used to treat small low-volume targets in the liver such as early-stage cancer tumors. Current research projects are working towards robotic tools to augment the surgeons for which real-time tracking of targets in the body is required. However, this is not possible with MRI alone due to the trade-off between resolution and imaging speed. Currently, an MRI of satisfactory quality requires four to five seconds of imaging time. Recent researches have used a sensor fusion approach to tackle this problem by combining the high resolution and contrast of an MRI with the update speed of a surrogate sensor like: ultrasound, IMUs, reference needles, visual markers, etc. allowing for interpolation between MRI scans and extrapolation outside the MRI bore.
This research will continue the work on combining MRI with surrogates using learning based algorithms and update it to work with state-of-the-art MRI formation. Next to this, work will be done on classifying and predicting respiratory motion for near-future target motion prediction. MRI scans will be performed on subjects with several surrogate sensors in order to create a versatile database suitable for the performance assessment of the sensor fusion, respiratory motion classification and its prediction.