Optical Flow based Interpolation of Temporal Image Sequences
Introduction
Modern tomographic imaging devices enable the acquisition of temporal image sequences and the study of organ motion becomes more and more important. For example cine MRI is used for the functional analysis of the heart and 4D-CT data sets are used for modelling organ motion during the respiratory cycle. But, in general, the spatial and temporal resolution of imaging devices is limited and a compromise between spatial resolution, temporal resolution, acquisition time and signal to noise ratio must be found. Therefore, in a number of image processing tasks a spatial and temporal interpolation of data sets is necessary to calculate dense motion models for instance. In our project, we compare cardiac cine MRI sequences of different patients, acquired with different temporal resolutions. A temporal interpolation of the image data is necessary to generate images at predefined phases of the cardiac cycle.
Methods
In this project methods for the interpolation of temporal image sequences were developed. We derive our interpolation scheme from the optical flow equation. The presented interpolation method consists of two steps: First the time-dependent optical flow field is determined using a non-linear registration method. Here, an iterative algorithm is applied, using the spatial and temporal image derivatives and a spatiotemporal smoothing step. Following the calculated optical flow field is used to generate
interpolated images for arbitrary time points. An interpolated image is generated at the desired time by averaging intensities between corresponding points.
Results
The behaviour and capability of the algorithm is demonstrated by synthetic image examples. In an evaluation procedure we calculated quantitative measures to compare our interpolation method with two other methods: linear and shape-based interpolation. Linear interpolation is the most frequently used interpolation technique and was chosen as a baseline reference. The shape-based interpolation algorithm was chosen since it was shown to have the best performance in a comparison of interpolation methods.
For the quantitative evaluation five cardiac MRI datasets (ECG-triggered true FISP sequences, between 13 and 21 time frames, 224 _ 256 pixels) were used (see Fig.1).
Results indicate that the presented method outperforms both linear and shape-based interpolation significantly.
Figure 1. Top row: Three consecutive slices of a cine MR sequence. The interpolation results for the second slice are shown in the middle row. Middle row: Sample slice estimated by linear (left), optical flow based (middle) and shape-based (right) interpolation. Bottom row: Corresponding difference images between the interpolated and original slice. The linear interpolated image (left) appears blurred and large differences can be observed. The shape-based interpolation (right) conserves edges of image structures but small details are lost. The optical flow-based interpolation (middle) performs more accurately and only few differences are shown in the difference image.
Project Team
Jan Ehrhardt
Dennis Säring
Heinz Handels
Selected Publications
Jan Ehrhardt, Dennis Säring, Heinz Handels: Interpolation of Temporal Image Sequences by Optical Flow based Registration. In: Reinhardt, J.M., Pluim, P.W. (eds.), Image Processing, SPIE Medical Imaging 2006, San Diego, Vol. 6144, 61442K-1-61442K-8, 2006
Jan Ehrhardt, Dennis Säring, Heinz Handels: Optical Flow-based Interpolation of Temporal Image Sequences. In: Handels, H., Erhardt, J., Horsch, A., Meinzer, H.P., Tolxdorff, T. (Hrsg.), Bildverarbeitung für die Medizin 2006, Informatik aktuell, Springer Verlag, Berlin, 256-260 2006.
