CT data completion based on prior scans

CT often suffers from missing data. In clinical CT so-called hollow projections arise when metal objects cannot be sufficiently penetrated by the x-rays. In flat detector CT additional artifacts originate from projection truncation, which occurs when the patient is larger than the field of measurement covered by the detector. Other artifacts arise from a limited scan angle, which is the case in some interventional C-arm CT devices which cannot perform a full 1800 rotation. By now, each of these artifacts have been approached using specific correction methods performing data interand extrapolation. The existing correction methods often struggle with completely removing the artifacts while not introducing new artifacts at the same time. In many cases, such as in interventional CT or image-guided radiation therapy, planning CT data are available that might serve as prior knowledge to correct for the missing CT data. Even if no such planning CT data are available, one may benefit from using prior information from a patient database where a similar patient may serve as the data source. To investigate this option we propose a generalized correction method capable of correcting all kinds of artifacts resulting from missing data by using prior knowledge of, e.g., a planning-CT scan or even data of another patient. The enabling technology allowing for an inter-patient correction method is a registration algorithm, followed by forward projection, smooth sinogram inpainting and image reconstruction. While being aware that these prior data can also be used to correct for spectral effects, such as beam hardening, or for scattered radiation, we currently restrict ourselves to correct for data that are completely missing. Our results indicate that many types of artifacts resulting from missing data can be corrected for effectively by using adequate prior data to perform data completion. This is exemplarily illustrated by correcting metal artifacts in a thorax scan, by correcting - or truncated CT data in a hip scan, and by correcting artifacts due to a limited scan angle in a head scan.