In my current job I spend my time training vision AI models and building complex data pipelines to solve tasks otherwise done by the human eye. Think counting, reading or search tasks. On top of that I’m working on my project management and Python programming skills.

When a patient moves during the acquisition of a CT scan, motion artifacts appear in the resulting images. These artifacts make it more difficult for radiologists and algorithms to analyse the scans. Detection networks can be used to identify affected areas and flag 'bad' scans in automated pipelines. This could lead to improved algorithm performance and diagnosis quality. Training such a detection network requires large amounts of annotated CT data, however, manual annotations can be subjective, unreliable, and are labour intensive to obtain.In this internship I developed a new method for motion artifact generation and an improved approach for motion artifact detection. Artificially generating artifacts especially increases the performance of detection networks when the availability of annotated CT data is limited.

Blind people have the possibility to see again through phosphene vision. Using small electrical stimulations of the visual cortex a pattern of small light flashes(phosphenes) can be observed without the need of eyes. However, the complexity of such phosphene patterns is limited. Therefore transforming a complex visual environment into a simplified phosphene stimulation protocol is a big challenge. During this internship I worked under the guidance of Jaap de Ruyter van Steveninck to find out whether we can tackle this problem using deep learning and build a data pipeline that transforms camera images into a meaningful phosphene stimulation protocol.