Lower extremity CT angiography (CTA): initial evaluation of a knowledge-based centerline estimation algorithm for femoro-popliteal artery (FPA) occlusions.

RATIONALE AND OBJECTIVES: Existing density- and gradient-based automated centerline-extraction algorithms fail in severely diseased or occluded arterial segments for the generation of curved planar reformations (CPRs). We aimed to quantitatively and qualitatively assess the precision of a knowledge-based centerline-extraction algorithm in patients with occluded femoro-popliteal artery (FPA). MATERIAL AND METHODS: Computed tomography angiograms of 38 FPA occlusions (mean length 120 mm) were retrospectively identified. Reference centerlines were determined as the mean of eight manual expert readings. Each occlusion was also interpolated using a new knowledge-based algorithm (partial vector space projection [PVSP]), which uses shape information extracted from a separate database of 30 nondiseased FPAs. Precision of PVSP was quantified as the maximum departure error (MDE) from the standard of reference and the proportion of the interpolated centerlines remaining within an assumed vessel radius of 3 mm. Multiple regression method was used to determine the factors predicting the precision of the algorithm. CPR quality was independently assigned by two readers. RESULTS: The mean MDE (in mm) for occlusion lengths of <50 mm, 50-100 mm, 100-200 mm, and >200 mm was 0.95, 1.19, 1.40, and 2.25, for manual readings and 1.68, 2.90, 9.43, and 19.95 for PVSP, respectively. MDEs of the algorithm were completely contained within 3 mm of the assumed vessel radius in 20 of 38 occlusions. CPR quality was rated diagnostic by both readers in 23 of 38 occlusions. CONCLUSION: Shape-based centerline extraction of FPA occlusions in lower extremity CTA is feasible, and independent from local density and gradient information. PVSP centerline extraction allows interpolation of occlusions up to 100 mm within the variability of manually derived centerlines.

Dual-energy CT discrimination of iodine and calcium: experimental results and implications for lower extremity CT angiography.

RATIONALE AND OBJECTIVES: The purpose of this work was to measure the accuracy of dual-energy computed tomography for identifying iodine and calcium and to determine the effects of calcium suppression in phantoms and lower-extremity computed tomographic (CT) angiographic data sets. MATERIALS AND METHODS: Using a three-material basis decomposition method for 80- and 140-kVp data, the accuracy of correctly identified contrast medium and calcium voxels and the mean attenuation before and after calcium suppression were computed. Experiments were first performed on a phantom of homogenous contrast medium and hydroxyapatite samples with mean attenuation of 57.2, 126, and 274 Hounsfield units (HU) and 50.0, 122, and 265 HU, respectively. Experiments were repeated in corresponding attenuation groups of voxels from manually segmented bones and contrast medium-enhanced arteries in a lower-extremity CT angiographic data set with mean attenuation of 293 and 434 HU, respectively. Calcium suppression in atherosclerotic plaques of a cadaveric specimen was also studied, using micro-computed tomography as a reference, and in a lower-extremity CT angiographic data set with substantial below-knee calcified plaques. RESULTS: Higher concentrations showed increased accuracy of iodine and hydroxyapatite identification of 87.4%, 99.7%, and 99.9% and 88.0%, 95.0%, and 99.9%, respectively. Calcium suppression was also more accurate with higher concentrations of iodine and hydroxyapatite, with mean attenuation after suppression of 47.1, 122, and 263 HU and 7.14, 11.6, and 12.6 HU, respectively. Similar patterns were seen in the corresponding attenuation groups of the contrast medium-enhanced arteries and bone in the clinical data set, which had overall accuracy of 81.3% and 78.9%, respectively, and mean attenuation after calcium suppression of 254 and 73.7 HU, respectively. The suppression of calcified atherosclerotic plaque was accurate compared with the micro-CT reference; however, the suppression in the clinical data set showed probable inappropriate suppression of the small vessels. CONCLUSION: Dual-energy computed tomography can detect and differentiate between contrast medium and calcified tissues, but its accuracy is dependent on the CT density of tissues and limited when CT attenuation is low.

Femoropopliteal artery centerline interpolation using contralateral shape.

Curved planar reformation allows comprehensive visualization of arterial flow channels, providing information about calcified and noncalcified plaques and degrees of stenoses. Existing semiautomated centerline-extraction algorithms for curved planar reformation generation fail in severely diseased and occluded arteries. We explored whether contralateral shape information could be used to reconstruct centerlines through femoropopliteal occlusions. We obtained CT angiography data sets of 29 subjects (16m/13f, 19-86yo) without peripheral arterial occlusive disease and five consecutive subjects (1m/4f, 54-85yo) with unilateral femoropopliteal arterial occlusions. A gradient-based method was used to extract the femoropopliteal centerlines in nondiseased segments. Centerlines of the five occluded segments were manually determined by four experts, two times each. We interpolated missing centerlines in 2475 simulated occlusions of various occlusion lengths in nondiseased subjects. We used different curve registration methods (reflection, similarity, affine, and global polynomial) to align the nonoccluded segments, matched the end points of the occluded segments to the corresponding patent end points, and recorded maximum Euclidean distances to the known centerlines. We also compared our algorithm to an existing knowledge-based PCA interpolation algorithm using the nondiseased subjects. In the five subjects with real femoropopliteal occlusions, we measured the maximum Euclidean distance and the percentage of the interpolation that remained within a typical 3 mm radius vessel. In the nondiseased subjects, we found that the rigid registration methods were not significantly (p<0.750) different among themselves but were more accurate than the nonrigid methods (p<0.001). In simulations using nondiseased subjects, our method produced centerlines that stayed within 3 mm of a semiautomatically tracked centerline in occlusions up to 100 mm in length; however, the PCA method was significantly more accurate for all occlusions lengths. In the actual clinical cases, we found the following [occlusion length (mm):error (mm)]: 16.5:0.775, 42.0:1.54, 79.9:1.82, 145:3.23, and 292:6.13, which were almost always more accurate than the PCA algorithm. We conclude that the use of contralateral shape information, when available, is a promising method for the interpolation of centerlines through arterial occlusions.