Standard MRI data is mostly in a qualitative form (showing the contrast between different tissues). Electrical properties tomography (EPT MRI) is a quantitative technique, which means that it shows a specific value for each voxel. In this thesis we researched the field of EPT MRI and found that machine learning can be used to detect the presence of an anomaly in EPT MRI images. EPT MRI brain images of phantoms were used as input data. As part of the thesis, we wrote two versions of the anomaly detection algorithm. The classical approach for anomaly detection detects regions, separated by edges in white matter, and then determines which regions are an anomaly, based on the mean value of the electric conductivity. The classical approach has detected anomalies that are approximately the size of a cube with 14 mm long edges. The second version of the algorithm exploits the quantitative characteristics of EPT MRI and detects anomalies through absolute values of electrical conductivity. We have detected anomalies that are approximately the size of a cube with 12 mm long edges via absolute values.
|
