Azimuthal registration of image sequences affected by NURD

A new algortihm to correct for Non-Uniform Rotation Distortion (NURD) has been developed and successfully implemented.

NURD
Imaging modalities that use a mechanically rotated endoscopic probe to scan a tubular volume, such as an artery, often suffer from image degradation due to non-uniform rotation distortion (NURD). Intravascular OCT is a particularly challenging modality for motion correction, because the image is generally undersampled, and correlation between the speckle in different lines or frames is absent.

Dynamic time warping
We have developed a new method to align individual lines in a sequence of images. It is based on dynamic time warping (DTW), finding a continuous path through a cost matrix that measures the similarity between regions of two frames being aligned. The path represents the angular mismatch corresponding to the NURD. The prime advantage of this novel approach compared to earlier work is the line-to-line continuity, which accurately captures intra-frame variations in rotational velocity of the probe. The algorithm can be adapted to ingest data frame by frame, and can be implemented to work in real time.

Clinical data
The algorithm was optimized using data from a clinically available intravascular Optical Coherence Tomography (OCT) instrument in a realistic vessel phantom. We demonstrate its efficacy on an in vivo recording, and compare it with conventional global rotation block matching. In the paper Van Soest et al., IEEE Trans. Inf. Technol. Biomed. 12(3), pp. 348-355 (2008) we present the technique in detail.

Examples
The effect of the NURD correction algorithm is most effectively visualized with actual movies. Two SWF clips demonstrating the method, in vitro and in vivo, are presented here as supplementary material for the paper. The in vitro images in this paper were acquired in a cryogel vessel phantom, using 312 lines per frame, 760 samples per line. The frame rate was 10 frames per second (fps). The movie shown below consists of 32 frames (looped). Left: the unprocessed image sequence; right: after DTW correction of NURD.


In vivo OCT movies were recorded in a male patient, age 51, in a healthy section of coronary artery. After inflation of an occlusion balloon the blood was cleared from the vessel using a saline flush. The frame rate was 15.6 fps and the frame size 200 lines by 744 pixels. The movie shown below consists of 50 frames (looped). Left: the unprocessed image sequence; middle: after correction with global rotation block matching; right: after DTW correction of NURD.
 


Note that the longitudinal motion of the catheter—arising because of cardiac motion—did not prevent the algorithm from correctly aligning the sequence. Features, such as the small side branch seen at 7 o'clock in the DTW aligned sequence, moved in and out of view with the cardiac cycle without changing their angular position. Neither did a small in-plane translation of the catheter that could be seen, or other cardiac motion, hamper the alignment.