Hybrid Feature Point Based Registration of 2D Abdominal CT Images
Liver lesions like abscess, cirrhosis, metastases etc usually
appear globular with gray levels differing remarkably from that
of the surrounding liver region but the boundaries of these
lesions are poorly defined and give an approximate picture of the
extent of invasion of the lesion into surrounding regions. Choice
of feature points and their matching in the images under
consideration is crucial and image dependent. This leads to
variations in registration methods. The objective of the proposed
technique is to enhance registration for such abdominal CT
images with the use of invariant points lying on the boundaries
of lesions and the lesion centroids as the feature points. This
paper uses a combination of hull points and centroids as feature
points to register the images and compares results with those
obtained by registration using only centroids as feature points.
The outer margins of lesions become more sharply defined which
helps in improving diagnostic accuracy.
Keywords: image registration, elastic registration, feature
point, wavelet decomposition
Download Full-Text
