Master thesis defense by Evangelos Vouros

Reducing the Need for General Anesthesia in Children Undergoing Neuroimaging by Preparation and Motion Correction

In Magnetic Resonance Imaging (MRI) of the human brain, the most common cause of artifacts generation is head motion. This problem is more severe when it comes to scanning children. This is because, contrary to adult patients who have in general better control over their movements, pediatric patients have a harder time doing so. The size of the scanner combined with the loud noises make this situation even worse. As a result, doctors usually resort in administering General Anesthesia (GA) to the children to get a diagnosable image. This however entails many health risks and financial problems because of the sedation through drugs and the additional personnel needed. In this project we came up with a method to avoid those problems and still get good images. Instead of sedating the children, we trained them to become familiar with the whole experience of getting inside a massive scanner and thus be calmer during the actual scan while, at the same time, they had the opportunity to watch a movie and be distracted from the loud noises. We also used a motion correction device which would, up to a certain degree, correct the motion artifacts of the image. However, this technique was applied on approximately half of the
children, chosen randomly, in an attempt to check whether it was the training or the motion correction that was more effective. The protocols consisted of T1 and T2 weighted, 2D or 3D encoded sequences and diffusion weighted imaging, depending on the structures and the parts of the brain that we wanted to focus on respectively. As part of our evaluation we tried to find a reliable way to assess the quality of the images by implementing image quality metrics that can produce a result without using a second image serving as a ground truth, since there is only one image provided in real clinical scans. Additionally, we studied
how accurately some metrics based on external motion measurements predict the image quality. These methods were first tested on data acquired from healthy volunteer patients and they were afterwards applied on the data from the pediatric patients. All those metrics were compared with quality scores based on the Likert scale given by doctors. The final
results of the thesis indicate that the image quality metrics showed a weak correlation with the results provided by the doctors while those depending on motion measurements turned out to be more reliable. Furthermore, we observed a significant decrease in the motion during the acquisition before and after the training, meaning that our initial assumption
was correct, having the children undergo a training before getting scanned can put out the need for sedation.