Artefact reduced Reconstruction of 4D-CT Data Sets from 3D-CT Data Segments

Problem

Modern CT scanners are able to acquire multiple slices at the same time. But the volume scanned within one rotation of the gantry is very often too small to image organ motion. But applications e.g. in radiation oncology may need information about organ motion. It is for example important to image and measure organ motion caused by breathing. For this reason tumour patients have been scanned several times with a CT scanner delivering single 3D-CT data segments. With the use of these segments 4D-CT data sets can be reconstructed. They can be used for further analysis of organ motions due to breathing.

Methods

A modern multi-slice CT scanner has been used to scan a part (= segment) of the patients during free breathing [1]. The position within the breathing cycle was measured with a spirometer so that multiple CT data segments were acquired at different positions within the breathing cycle. By moving the CT couch data segments for the whole thoracic region could be scanned.

The 3D-CT data segments have to be processed to reconstruct a 4D-CT data set from the 3D-CT data segments. Therefore data segments are needed for each couch position that have been acquired in the same position within the breathing cycle. Using standard techniques, segments are selected that are closest to the lung volume chosen (fig. 1). This method can be called "nearest neighbour interpolation". Segments that are most similar to the volume desired are chosen. By the use of this method artefacts occur that are visible especially at the diaphragm.

To reduce these artefacts an interpolation method has been developed using non-linear image registration methods. The non-linear registration method is based on the "optical flow". The application of this method on two data segments approximates a vector transformation field describing the shift of the single voxels between neighbouring lung volumes. With this 3D transformation field data segments can be calculated for the lung volume selected. The results (fig. 2) show less artefacts than the images reconstructed with the nearest neighbour strategy (fig. 1).