Medical imaging datasets have achieved significant advancements in resolution and detail. Despite the efforts of practitioners to anonymize the data, patients might be still recognizable due to the high-level-of-detail acquisitions. Often, anonymization of volumetric data limits itself with removing a region of interest (e.g., the face from a head data set), or adding noise and other effects. Conversely, image inpainting has proven to be successful in anonymizing, for example, camera footage (1,2,3).
In this project, we want to investigate whether inpainting techniques can be applied to the anonymization of volume rendering. Inpainting in volume rendering can be challenging, as (i) the underlying medical data should not be altered---only the characteristics that give away the identity of the patient, and (ii) high resolution should be retained without causing disruptions and distortions in the aesthetics of the volume representation.
- Interest and knowledge in volume rendering.
- Interest and/or knowledge in image/video inpainting.
- Good programming skills.
- Creativity and enthusiasm.
To be discussed (depending on the background of the student).