Session PC7.3

A Novel Approach to Quantitative Analysis of Intravascular Optical Coherence Tomography Imaging

K Sihan*, C Botha, F Post, S de Winter,
E Regar, R Hamers, N Bruining

Technical University
Delft, Netherlands

Quantitative Coronary analysis on intravascular optical coherence tomography (OCT) data (QOCT) is currently performed by manual contour tracing in cross-sectional images of OCT pullback procedures (frame-based method). For a comprehensive three-dimensional (3D) assessment of coronary dimensions of a long segment, analyses derived from many cross-sectional areas, were different contours of corresponding structures need to be traced, results in a time-consuming procedure. Furthermore, the OCT data is acquired non-gated resulting in a saw-tooth shaped appearance of the coronary vessel wall, making it difficult or even impossible to use longitudinal views (L-views) for contour tracing.
In order to get a more efficient analysis procedure and to investigate if image-based retrospective OCT gating would be possible, as first step a novel approach has been developed that exploits a fully automatic contour tracing method for coronary lumens in OCT images.
The OCT images are first translated into the DICOM imaging standard format and pre-processed needing a minimum interaction of the users, e.g. the user must identify the center of the catheter. After this the images are filtered with a median filter (to get rid of possible displayed grids into the images), followed by a Gaussian (to get rid of noise) and a Wiener filter (also to get rid of noise). To get rid of black holes a maximum and a minimum filter are than applied. After these pre-processing steps contour detection is performed by applying ray-casting. The so detected contours are re-examined by a 3D quality check algorithm, were first the images with a high probability of correct contours are identified, after which the contours with lower probabilities are checked. In case side-branches or the vessel is out of the image plane (or other image artifacts) is encountered, contour information from adjacent contours with a high probability are interpolated towards these lower probability contours.
The contours are than finally transferred to snakes, which can be easily enhanced by expert human observers if necessary.
Automated contour detection of lumen contours in OCT images is investigated showing promising results. Future work must address if the so derived contour information could be used for image-based retrospective gating for OCT.

(Abstract Control Number: 281)