Title :
Vascular architecture in subharmonic breast images: A comparative study
Author :
Forsberg, Flemming ; Joshi, Neha ; Eisenbrey, John R.
Author_Institution :
Dept. of Radiol., Thomas Jefferson Univ., Philadelphia, PA, USA
Abstract :
Three algorithms for extracting vascular architecture in contrast-enhanced subharmonic imaging (SHI) were compared. Skeletonization of binary images was achieved by applying either sequential or parallel thinning or a non-iterative method using the distance transformation. The 2 iterative algorithms conditionally erode the vessel using local neighborhood patterns as sufficient condition, while the non-iterative method calculates the center line of the vessel directly. Results from 40 test images (letters and numbers) were compared based on data reduction efficiency, medial axis representation, connectivity and rotational invariance. In vivo grayscale SHI data was obtained from 16 breast lesions using a modified Logiq 9 scanner (GE Healthcare, Milwaukee, WI) and Cumulative Maximum Intensity (CMI) SHI images, which retain maximum intensity pixels from successive frames, were created. Lesion types were analyzed in terms of bifurcations and average vessel-chain length of the vascular architecture. All algorithms retained connectivity in vitro, but only the sequential algorithm was rotational invariant. The data reduction efficiency was better with sequential thinning (0.93+0.01) than with the other 2 methods (0.82+0.02 and 0.90+0.02 for parallel and distance transformation, respectively; p<;0.001). For medial axis representation distance transformation did best (p<;0.01). In vivo, vascular skeletonization with the sequential algorithm was achieved for 13 of 16 lesions (due to lack of vasculature in 3 cases). Lesions showed an average bifurcation number of 7.6+6.78 and an average vessel-chain length of 71.1+55.73 pixels/chain. Malignant lesions had a higher number of bifurcations compared to benign lesions (9.8+8.18 vs. 6.9+6.50) and a larger average vessel-chain length (88.9+79.10 vs. 63.2+45.65 pixels/chain), although these differences were not statistically significant (p>;0.4). In conclusion, 3 different algorithms for skeletonizing the vascular structure- - s of breast lesions depicted in SHI mode were compared. Sequential thinning produced the best results in vitro and in vivo.
Keywords :
biomedical ultrasonics; blood vessels; medical image processing; ultrasonic imaging; Cumulative Maximum Intensity SHI images; average vessel-chain length; bifurcations; binary image skeletonization; breast lesions; connectivity; contrast-enhanced subharmonic imaging; data reduction efficiency; distance transformation; medial axis representation; modified Logiq 9 scanner; noniterative method; parallel thinning; rotational invariance; subharmonic breast images; vascular architecture; Breast; Imaging; In vitro; In vivo; Lesions; Pixel; Ultrasonic imaging; breast imaging; image processing; subharmonic signals; ultrasound contrast agents;
Conference_Titel :
Ultrasonics Symposium (IUS), 2010 IEEE
Conference_Location :
San Diego, CA
Print_ISBN :
978-1-4577-0382-9
DOI :
10.1109/ULTSYM.2010.5935799