Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online before print February 20, 2007, 10.1148/radiol.2431060041
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2431060041v1
243/1/56    most recent
Right arrow Submit a response
Right arrow View responses
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Chang, R.-F.
Right arrow Articles by Chen, D.-R.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Chang, R.-F.
Right arrow Articles by Chen, D.-R.
(Radiology 2007;243:56-62.)
© RSNA, 2007

Breast Imaging

Solid Breast Masses: Neural Network Analysis of Vascular Features at Three-dimensional Power Doppler US for Benign or Malignant Classification1

Ruey-Feng Chang, PhD, Sheng-Fang Huang, PhD, Woo Kyung Moon, MD, Yu-Hau Lee, MS and Dar-Ren Chen, MD

1 From the Department of Computer Science and Information Engineering and Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei, Taiwan (R.F.C.); Department of Medical Informatics, Tzu Chi University, Hualien, Taiwan (S.F.H.); Department of Computer Science and Information Engineering, National Chung Cheng University, Chiayi, Taiwan (Y.H.L.); Department of Radiology and Clinical Research Institute, Seoul National University Hospital and the Institute of Radiation Medicine, Seoul National University Medical Research Center, 27 Yongon-dong, Chongno-gu, Seoul 110-744, Korea (W.K.M.); and Department of Surgery, Changhua Christian Hospital, Changhua, Taiwan (D.R.C.). Received January 9, 2006; revision requested March 9; revision received April 3; accepted May 9; final version accepted August 1. Address correspondence to W.K.M. (e-mail: moonwk{at}radcom.snu.ac.kr).


   ABSTRACT
TOP
 ABSTRACT
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 
Purpose: To retrospectively evaluate the accuracy of neural network analysis of tumor vascular features at three-dimensional (3D) power Doppler ultrasonography (US) for classification of breast tumors as benign or malignant, with histologic findings as the reference standard.

Materials and Methods: This study was approved by the local ethics committee; informed consent was waived. Three-dimensional power Doppler US images of 221 solid breast masses (110 benign, 111 malignant) were obtained in 221 women (mean age, 46 years; range, 25–71 years). After narrowing down vessels to skeletons with a 3D thinning algorithm, six vascular feature values—vessel-to-volume ratio, number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were computed. A neural network was used to classify tumors by using these features. Independent-samples t test and receiver operating characteristic (ROC) curve analysis were used.

Results: Mean values of vessel-to-volume ratio, number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter were 0.0089 ± 0.0073 (standard deviation), 26.41 ± 14.73, 23.02 cm ± 19.53, 8.44 cm ± 10.38, 36.31 ± 37.06, and 0.088 cm ± 0.021 in malignant tumors, respectively, and 0.0028 ± 0.0021, 9.69 ± 6.75, 5.17 cm ± 4.78, 1.68 cm ± 1.79, 6.05 ± 7.55, and 0.064 cm ± 0.028 in benign tumors, respectively (P < .001 for all six features). Area under ROC curve (Az) values of the six features were 0.84, 0.87, 0.87, 0.82, 0.84, and 0.75, respectively. Accuracy, sensitivity, specificity, and positive and negative predictive values were 85% (187 of 221), 83% (96 of 115), 86% (91 of 106), 86% (96 of 111), and 83% (91 of 110), respectively, with Az of 0.92 based on all six feature values.

Conclusion: Three-dimensional power Doppler US images and neural network analysis of features can aid in classification of breast tumors as benign or malignant.

© RSNA, 2007

The potential contribution of two-dimensional or three-dimensional (3D) power Doppler ultrasonography (US), with or without contrast agents, for characterization of breast tumors has been stressed in some reports (1–7), but the issue remains controversial because benign and malignant tumors may be vascular. Thus, the detection of tumor vessels alone may be insufficient to accurately differentiate benign from malignant breast masses (2). As opposed to blood vessels in normal tissues or benign tumors, blood vessels in cancerous tissues or tumors do not follow an ordered growth. Instead, tumor angiogenesis results in excessive dilatation and branching and tortuous patterns (8).

Subjective observations confirm the relatively chaotic vascular geometry in malignant tumors compared with that in benign tumors (3,9). As such, morphologic features are likely to be additional clues, which when used in conjunction with more established parameters such as vessel density or flow velocity can improve the present diagnostic approaches. Quantitative analysis of tumor vascularity presents a challenge, and the majority of previous analytic studies (10,11) have focused on vessel amounts based on pixel or voxel counts rather than based on vessel morphology.

For quantitative vascular morphologic analysis on medical images, a sequential procedure of 3D image formation, preprocessing, segmentation, thinning, skeleton pruning, and tree construction is required (12,13). Quantitative data, such as vessel length, diameter, and bifurcation angle, may be acquired from vascular trees composed of a set of nodes and connecting arcs. Three-dimensional power Doppler US offers a means of reconstructing the vascular architecture, and computer-aided analysis could be applied to extract information about tumor vascularity (14). To our knowledge, neural network analysis of vascular features on 3D power Doppler US images has not been used for classification of breast tumors as benign or malignant. Thus, the purpose of our study was to retrospectively evaluate the accuracy of a neural network analysis of vascular features at 3D power Doppler US for classification of breast tumors as benign or malignant, with histologic findings as the reference standard.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 
Patients
This study was approved by the local ethics committee, and informed consent was waived for our retrospective study. Between January and March 2003, 246 consecutive women who had been scheduled to undergo excisional or percutaneous needle biopsy on the basis of suspicious mammographic or US findings underwent 3D power Doppler US. A total of 221 solid breast masses in 221 women (mean age, 46 years; age range, 25–71 years) definitely visualized with tumor vessels at 3D power Doppler US were included in this study. The remaining 25 patients with solid breast masses were excluded because vessels associated with the tumor mass were not visualized at 3D power Doppler US (Fig 1).


Figure 1
View larger version (16K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1: Flow diagram of study patients.

 
The database used in our study contained 110 pathologically proved (reference standard) benign tumors (59 fibroadenomas, 51 fibrocystic lesions) and 111 pathologically proved invasive carcinomas (107 invasive ductal carcinomas, three lobular carcinomas, one medullary carcinoma). The size of the benign tumors was 0.5–3.0 cm (mean, 1.5 cm), whereas the size of the malignant tumors was 0.9–3.7 cm (mean, 1.7 cm). All 221 lesions had been initially classified by attending radiologists as Breast Imaging Reporting and Data System category 3 (15), probably benign lesions (n = 16); category 4, suspicious lesions (n = 127); or category 5, highly suspicious lesions (n = 78). All patients with malignant lesions underwent surgery within 2 weeks (range, 6–14 days; mean, 9.7 days; median, 11.0 days) of US examination. Mammographic and sonographic follow-up was performed in 76 (69.1%) benign lesions, and mean duration of follow-up was 11.2 months (range, 6–24 months). In the other 34 (30.9%) benign lesions, imaging follow-up was not performed at our institution.

US Imaging
All US images were acquired with a 3D power Doppler US scanner (Voluson 730; GE Kretz, Zipf, Austria) and a 5–10-MHz dedicated volume transducer by the two authors (W.K.M., D.R.C.; 13 and 10 years of experience, respectively, with breast US). The following settings were used: low wall filter, pulse repetition frequency of 0.9, and mechanical index of 0.5. A suitable volume box size was chosen to include the lesion and a minimal amount of normal surrounding tissue. The volume scans were automatically acquired by using a slow-tilt movement of the sectorial mechanical transducer. Three-dimensional volume files were saved in Cartesian coordinates by using the program with the US scanner (3D-View 2000; GE Kretz). The mean pixel resolution was 0.019 cm/pixel or 51.74 pixel/cm.

Image Preprocessing
A 3D power Doppler US data set is essentially an array of voxels represented in the red-green-blue color system. Image preprocessing was performed by three authors (R.F.C., S.F.H., Y.H.L.) to separate vascular points from the background. Original 3D power Doppler US images were converted into binary data by applying a predefined threshold value, TR, to examine the red channel for each voxel. A TR value of 120 was manually selected by experienced physicians (W.K.M., D.R.C.) and applied to all study images. For voxels, p, if the red component R was greater than TR, p was assigned a value of 0, and if not, it was assigned a value of 255. Thus, the voxels of value 0 on binary images were used to estimate the amounts of vascular points.

To compensate for cavities within vessels and to exclude artifacts caused by nonuniform luminance and noise, a sequence of digital morphologic operations, previously defined as an ordered combination of dilation and erosion (16), was performed on acquired binary images prior to capturing centerlines.

Three-dimensional Thinning Algorithm
To produce vessel skeletons in 3D discrete space, we used the 3D thinning algorithm proposed by Palagyi and Kuba (17). For 3D binary images, this thinning algorithm can directly create centerlines from an elongated or tubular object without the need to generate medial surfaces first. Centerlines are defined as a set of 1-pixel-wide voxel lists that represent the skeleton of vessels. Each iteration step involved six successive subiterations that can be executed in parallel. That is, points of the targeted object on a binary image that satisfy the deletion requirements can be changed to background simultaneously. "Skeletonization," or extraction of the 3D centerline, is a common way of representing shape features by using a limited amount of binary image data (13,18). Skeletons can serve as one-dimensional structures and can be used for the orderly exploration of the entire tree and can simplify the task of volume manipulation.

Vascular Tree Construction
To generate a skeletal representation of vascular structures, three authors (R.F.C., S.F.H., Y.H.L.) converted centerlines into a tree structure consisting of a set of nodes and connecting arcs by using the "breadth first" search algorithm (19). Localization of the root is specified by using the voxel detected in the first nonzero two-dimensional section in the z-axis. Every voxel in a given skeleton corresponds to a tree node, with edges pointing to other nodes; nodes having more than one successor were marked as bifurcations (Fig 2a). Important structural information was documented for further evaluation while the vascular trees were being recursively traversed.


Figure 2A
View larger version (12K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2a: Vascular tree construction. (a) A bifurcation (solid circle) is a node that has more than one child node. (b) At each bifurcation, any branch that contains only a single leaf node (dotted line with gray open circle) is considered to be pruned, while one that contains at least one child (solid line with open circle) is preserved (see node z). x = node x, y = node y.

 

Figure 2B
View larger version (11K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2b: Vascular tree construction. (a) A bifurcation (solid circle) is a node that has more than one child node. (b) At each bifurcation, any branch that contains only a single leaf node (dotted line with gray open circle) is considered to be pruned, while one that contains at least one child (solid line with open circle) is preserved (see node z). x = node x, y = node y.

 
Because most skeletonization algorithms are sensitive to noise and coarse boundaries, we performed a simple postprocessing step to prune out unwanted branching segments caused by false generation (Fig 2b) (20).

Feature Extraction
Values of six vascular features of tumors—vessel-to-volume ratio, total number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were computed by three authors (R.F.C., S.F.H., Y.H.L.) to evaluate the findings of benign and malignant tumors (8,21). The first feature, vessel-to-volume ratio, was determined as the number of color pixels divided by the number of total pixels in the volume of interest and was computed by using 3D power Doppler US images immediately after image preprocessing. The other five features were computed by using 3D power Doppler US images after 3D thinning and vascular tree reconstruction.

For feature extraction, we defined S as the set of centerlines produced containing n elements. Let S = {T1, T2, ..., Tn}. For x = 1, 2, ..., n, each element Tx denotes one of the pruned vascular trees. This can also be written as Tx (Vx, mx, Rx), where

Formula 1(1)
represents a set containing mx nodes, and Rx {epsilon} Vx indicates the root of Tx. Each node in Vx stores the coordinates of the corresponding voxel and a list of indices pointing to its successor nodes in the tree.

The vessel-to-volume ratio was previously determined after the stage of image thresholding, and because S = {Tx   1 ≤ x ≤ n} denotes the set of all vascular trees, the total number of vascular trees is n. Total vessel length was determined by summing up the length of edges for all Tx {epsilon} S, whereas the length of the longest path in S was computed recursively by using the algorithm proposed in reference 22. In our study, the evaluation of length was defined in terms of euclidean distance, and thus, we calculated the length of an edge between two adjacent nodes by using the true coordinates of corresponding voxels. For bifurcation, we counted the total number of bifurcate nodes for all vascular trees on the image to estimate tumor vessel branching.

Vessel diameter was obtained by setting twice the average primary path radius from centerlines. To obtain the radius at each voxel along centerlines, we overlapped the skeletons produced with the original binary image and reapplied the 3D thinning algorithm. The so-called primary path, Pt, is defined as a path in some tree Tx where x = 1, 2, ..., n, on which the sum of radiuses of all nodes in Pt are a maximum. The units of total vessel length, longest path length, and vessel diameter were adjusted to centimeter according to different voxel resolutions due to the various settings at US scanning.

Classification
A general multilayer perceptron neural network (23,24) with the back-propagation learning rule was used to classify solid breast tumors on the basis of the values of the six features. The values produced by the output node of the neural network lie between 0 and 1. We chose a threshold of 0.5 to classify the benign and malignant tumors after conducting experiments. Thus, if the output value was equal to or higher than 0.5, a tumor was classified as malignant. If the output value was less than 0.5, a tumor was regarded as benign. By using the k-fold cross-validation method (25), 220 (110 benign and 110 malignant tumors) of the 221 3D power Doppler US images in the database were divided randomly into five groups. One of 111 malignant lesions excluded was selected at random. One group (n = 44) consisted of 22 benign and 22 malignant tumors and was used as the test set. The remaining four groups were training sets.

The time required for thinning and vascular tree construction was 30–40 seconds per case. The computation time required for feature extraction and classification was 3–4 seconds per case. The total computation time required was 33–44 seconds per case.

Statistical Analysis
The mean values and standard deviations of the six features were calculated for benign and malignant tumors by using 3D power Doppler US images. Significant differences between the six feature values for benign and malignant tumors were evaluated with the independent-samples t test. Assumption of equal variances of the two populations was determined with the Levene test, and t test results then were interpreted accordingly (ie, Student t test when equal variances were assumed and Welch t test when the assumption was not fulfilled).

The performance of the values for these six features was evaluated by using a receiver operator characteristic (ROC) curve analysis program (LABROC1, 1993; Charles E. Metz, MD, University of Chicago, Chicago, Ill). Area under ROC curve (Az) results were used as indicators of performance. Values of one feature relating to the amount of vascularity—vessel-to-volume ratio—were compared with values of the five features relating to vessel morphology computed from 3D power Doppler US images after 3D thinning and tree reconstruction.

Diagnostic performance of the neural network on the basis of the values of the six features used for classification of solid breast tumors on 3D power Doppler US images was evaluated with accuracy, sensitivity, specificity, positive and negative predictive values, and Az analysis. For each analysis, a P value of less than .05 was considered to indicate a significant difference. Statistical analyses other than ROC analysis were performed with software (SPSS, version 10 for Windows; SPSS, Chicago, Ill).


   RESULTS
TOP
 ABSTRACT
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 
Features
The mean value of the feature relating to the amount of vascularity—vessel-to-volume ratio—was 0.0089 ± 0.0073 in malignant tumors and 0.0028 ± 0.0021 in benign tumors, and these values were significantly different (P < .001) (Figs 3, 4). The mean values of the five features relating to vessel morphology—number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were 26.41 ± 14.73, 23.02 cm ± 19.53, 8.44 cm ± 10.38, 36.31 ± 37.06, and 0.088 cm ± 0.021 in malignant tumors, respectively, and 9.69 ± 6.75, 5.17 cm ± 4.78, 1.68 cm ± 1.79, 6.05 ± 7.55, and 0.064 cm ± 0.028 in benign tumors, respectively. Differences between benign and malignant breast tumors were statistically significant for all five features (P < .001).


Figure 3A
View larger version (72K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3a: Three-dimensional power Doppler US images of malignant breast lesion. (a) Original data. (b) Original data after 3D thinning. (c) Data after vascular tree construction. In this case, values of six features of tumor vascularity—vessel-to-volume ratio, number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were 0.0182, 37, 34.325 cm, 17.096 cm, 79, and 0.085 cm, respectively.

 

Figure 3B
View larger version (51K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3b: Three-dimensional power Doppler US images of malignant breast lesion. (a) Original data. (b) Original data after 3D thinning. (c) Data after vascular tree construction. In this case, values of six features of tumor vascularity—vessel-to-volume ratio, number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were 0.0182, 37, 34.325 cm, 17.096 cm, 79, and 0.085 cm, respectively.

 

Figure 3C
View larger version (55K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3c: Three-dimensional power Doppler US images of malignant breast lesion. (a) Original data. (b) Original data after 3D thinning. (c) Data after vascular tree construction. In this case, values of six features of tumor vascularity—vessel-to-volume ratio, number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were 0.0182, 37, 34.325 cm, 17.096 cm, 79, and 0.085 cm, respectively.

 

Figure 4A
View larger version (59K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4a: Three-dimensional power Doppler US images of benign breast lesion. (a) Original data. (b) Original data after 3D thinning. (c) Data after vascular tree construction. In this case, values of six features of tumor vascularity—vessel-to-volume ratio, number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were 0.0030, 13, 6.193 cm, 1.133 cm, 2, and 0.061 cm, respectively.

 

Figure 4B
View larger version (12K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4b: Three-dimensional power Doppler US images of benign breast lesion. (a) Original data. (b) Original data after 3D thinning. (c) Data after vascular tree construction. In this case, values of six features of tumor vascularity—vessel-to-volume ratio, number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were 0.0030, 13, 6.193 cm, 1.133 cm, 2, and 0.061 cm, respectively.

 

Figure 4C
View larger version (14K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4c: Three-dimensional power Doppler US images of benign breast lesion. (a) Original data. (b) Original data after 3D thinning. (c) Data after vascular tree construction. In this case, values of six features of tumor vascularity—vessel-to-volume ratio, number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were 0.0030, 13, 6.193 cm, 1.133 cm, 2, and 0.061 cm, respectively.

 
ROC Analysis
The Az value for the single vascularity amount feature—vessel-to-volume ratio—was 0.84, whereas Az values for the five vascular morphologic features—number of vascular trees, total vessel length, longest path length, number of bifurcations, and vessel diameter—were 0.87, 0.87, 0.82, 0.84, and 0.75, respectively. Az values of two vascular morphologic features—number of vascular trees and total vessel length—were significantly higher than the Az value for the single vascularity amount feature (P = .04 and P = .03, respectively) (Fig 5).


Figure 5
View larger version (28K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5: ROC curves for six features and for multilayer perceptron neural network (All features). Az values of two vessel morphologic features—number of vascular trees (Nv) and total vessel length (L1)—were significantly higher than Az value of vessel amount (P = .04 and P = .03, respectively). Az value for multilayer perceptron neural network with all six features produced in the best performance, with Az value of 0.92. Rv = vessel-to-volume ratio, L2 = longest path length, Bn = number of bifurcations, Dv = vessel diameter.

 
Accuracy
By using the multilayer perceptron neural network, the accuracy, sensitivity, specificity, and positive and negative predictive values were 85% (187 of 221), 83% (96 of 115), 86% (91 of 106), 86% (96 of 111), and 83% (91 of 110). The Az value of the multilayer perceptron neural network on the basis of the values of all six features for classification of solid breast tumors was 0.92 by using 3D power Doppler US images.


   DISCUSSION
TOP
 ABSTRACT
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
 References
 
Results of our study show that on the basis of image processing and analysis of tumor vascular features by using 3D power Doppler US images, the proposed neural network method can be successfully used to classify breast tumors as benign or malignant. The Az value of our neural network analysis method was 0.92 for all six features combined in equivocal cases, which required an interventional procedure to determine tumor status. The total computation time required, which included preprocessing, 3D thinning, and feature analysis, was less than 45 seconds per case.

We focused on designing a method of quantification that reflected the nature of vascular structures as depicted by using 3D power Doppler US because qualitative and quantitative analyses based on pixel counts or flow parameters are not superior to the subjective analysis of vessel morphology (regular vs irregular) by radiologists (3,7,9). Az values of two vessel morphologic features—number of vascular trees and total vessel length—were significantly higher than the value of the vessel-to-volume ratio. Specific features of malignant neovascularity include a lack of normal vessel tapering pattern, vessel tortuosity, excessive branching, and radial penetration into the cancer from margins (6). Tumor vessels are both structurally and functionally abnormal. Future development of 3D power Doppler programs should include 3D Doppler shift spectrum information and should be designed to create applications that are more compatible with clinical practice (26,27).

There were limitations in our study. Twenty-five patients with solid breast masses were excluded from the study because tumor vessels associated with these masses were not depicted at 3D power Doppler US. Contrast material–enhanced 3D power Doppler US might have better depicted the chaotic morphologic structure of tumor neovascularity. Previous Doppler US studies (4,7) revealed that echo enhancement by using a contrast agent can increase diagnostic accuracy in terms of differentiation of malignant from benign breast tumors. Involuntary patient motion or transducer movement during data acquisition can introduce artifacts at 3D power Doppler US (27). In our study, an automatic mechanical scanning system was used instead of the freehand sweeping method for data acquisition, and this improvement could possibly have affected the accuracy of our computer-aided diagnosis system. In our study, 3D gray-scale anatomic information was not taken into account. Because several 3D breast US computer-aided classification systems based on analysis of breast tumor shape or texture have been reported to have good diagnostic performance (28,29), adaptation of these features in the future will help refine the diagnostic accuracy of our system.

In conclusion, we propose a method that extracts features of tumor vascularity from 3D breast power Doppler US images and a neural network system that uses these features to classify solid breast tumors as benign or malignant. The results show that 3D power Doppler US images and computer-aided analysis can aid in the classification of benign and malignant breast tumors.


   ADVANCES IN KNOWLEDGE
TOP
 ABSTRACT
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
References
 


   FOOTNOTES
 

Abbreviations: Az= area under ROC curve • ROC = receiver operating characteristic • 3D = three-dimensional

Authors stated no financial relationship to disclose.

Author contributions: Guarantors of integrity of entire study, R.F.C., W.K.M.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, R.F.C., W.K.M.; clinical studies, R.F.C., W.K.M., D.R.C.; experimental studies, all authors; statistical analysis, R.F.C., S.F.H., W.K.M., Y.H.L.; and manuscript editing, R.F.C., S.F.H., W.K.M., Y.H.L.


   References
TOP
 ABSTRACT
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 

  1. Kook SH, Park HW, Lee YR, Lee YU, Pae WK, Park YL. Evaluation of solid breast lesions with power Doppler sonography. J Clin Ultrasound 1999; 27:231–237.[CrossRef][Medline]
  2. Birdwell RL, Ikeda DM, Jeffrey SS, Jeffrey RB. Preliminary experience with power Doppler imaging of solid breast masses. AJR Am J Roentgenol 1997; 169:703–707.[Abstract/Free Full Text]
  3. Schroeder RJ, Maeurer J, Vogl TJ, et al. D-galactose-based signal-enhanced color Doppler sonography of breast tumors and tumorlike lesions. Invest Radiol 1999; 34:109–115.[CrossRef][Medline]
  4. Forsberg F, Goldberg BB, Merritt CR, et al. Diagnosing breast lesions with contrast-enhanced 3-dimensional power Doppler imaging. J Ultrasound Med 2004; 23:173–182.[Abstract/Free Full Text]
  5. Carson PL, Moskalik AP, Govil A, et al. The 3D and 2D color flow display of breast masses. Ultrasound Med Biol 1997; 23:837–849.[CrossRef][Medline]
  6. Raza S, Baum JK. Solid breast lesions: evaluation with power Doppler US. Radiology 1997; 203:164–168.[Abstract/Free Full Text]
  7. Moon WK, Im JG, Noh DY, Han MC. Nonpalpable breast lesions: evaluation with power Doppler US and a microbubble contrast agent—initial experience. Radiology 2000; 217:240–246.[Abstract/Free Full Text]
  8. Less JR, Skalak TC, Sevick EM, Jain RK. Microvascular architecture in a mammary carcinoma: branching patterns and vessel dimensions. Cancer Res 1991; 51:265–273.[Abstract/Free Full Text]
  9. Stuhrmann M, Aronius R, Schietzel M. Tumor vascularity of breast lesions: potentials and limits of contrast enhanced Doppler sonography. AJR Am J Roentgenol 2000; 175:1585–1589.[Abstract/Free Full Text]
  10. Carson PL, Fowlkes JB, Roubidoux MA, et al. 3-D color Doppler image quantification of breast masses. Ultrasound Med Biol 1998; 24:945–952.[CrossRef][Medline]
  11. Kettenbach J, Helbich TH, Huber S, Zuna I, Dock W. Computer-assisted quantitative assessment of power Doppler US: effects of microbubble contrast agent in the differentiation of breast tumors. Eur J Radiol 2005; 53:238–244.[CrossRef][Medline]
  12. Chen Z, Molloi S. Automatic 3D vascular tree construction in CT angiography. Comput Med Imaging Graph 2003; 27:469–479.[CrossRef][Medline]
  13. Yi D, Hayward V. Skeletonization of volumetric angiograms for display. Comput Methods Biomech Biomed Engin 2002; 5:329–341.[CrossRef][Medline]
  14. Tyrrell JA, Mehadevan V, Tong RT, Brown EB, Jain RK, Roysam B. A 2-D/3-D model-based method to quantify the complexity of microvasculature imaged by in vivo multiphoton microscopy. Microvasc Res 2005; 70:165–178.[CrossRef][Medline]
  15. American College of Radiology. Breast imaging reporting and data system (BI-RADS). 3rd ed. Reston, Va: American College of Radiology, 1998.
  16. Giardina CR, Dougherty ER. Morphological methods in image and signal processing. Upper Saddle River, NJ: Prentice-Hall, 1988.
  17. Palagyi K, Kuba A. A 3D 6-subiteration thinning algorithm for extracting medial lines. Pattern Recognit Lett 1998; 19:613–627.
  18. Blum H. A transformation for extracting new descriptors of shape. In: Wathen-Dunn W, ed. Models for the perception of speech and visual form. Cambridge, Mass: MIT Press, 1967; 362–380.
  19. Lee CY. An algorithm for path connection and its applications. IEEE Trans Electron Comput 1961; EC-10(3):346–365.
  20. Palagyi K, Tschirren J, Hoffman EA, Sonka M. Quantitative analysis of pulmonary airway tree structures. Comput Biol Med 2006; 36(9):974–996.[CrossRef][Medline]
  21. Kurjak A, Kupesic S, Breyer B, Sparac V, Jukic S. The assessment of ovarian tumor angiogenesis: what does three-dimensional power Doppler add? Ultrasound Obstet Gynecol 1998; 12:136–146.[CrossRef][Medline]
  22. Megiddo N, Tamir A, Zemel E, Chandrasekaran R. An O(n log2 n) algorithm for the k-th longest path in a tree with applications to location problems. SIAM J Comput 1981; 10:328–337.
  23. Freeman JA, Skapura DM. Neural networks: algorithms, applications, and programming techniques. New York, NY: Addison-Wesley, 1992.
  24. Chang RF, Wu WJ, Moon WK, Chou YH, Chen DR. Support vector machines for diagnosis of breast tumors on US images. Acad Radiol 2003; 10:189–197.[CrossRef][Medline]
  25. Haykin S. Neural networks: a comprehensive foundation. 2nd ed. Upper Saddle River, NJ: Prentice-Hall, 1999.
  26. Pairleitner H, Steiner H, Hasenoehrl G, Staudach A. Three-dimensional power Doppler sonography: imaging and quantifying blood flow and vascularization. Ultrasound Obstet Gynecol 1999; 14:139–143.[CrossRef][Medline]
  27. Nelson TR, Pretorius DH, Hull A, Riccabona M, Sklansky MS, James G. Sources and impact of artifacts on clinical three-dimensional ultrasound imaging. Ultrasound Obstet Gynecol 2000; 16:374–383.[CrossRef][Medline]
  28. Sahiner B, Chan HP, Roubidoux MA, et al. Computerized characterization of breast masses on three-dimensional ultrasound volumes. Med Phys 2004; 31:744–754.[CrossRef][Medline]
  29. Chen DR, Chang RF, Chen WM, Moon WK. Computer-aided diagnosis for 3-dimensional breast ultrasonography. Arch Surg 2003; 138:296–302.[Abstract/Free Full Text]



This article has been cited by other articles:


Home page
 J Ultrasound MedHome page
J. Du, F.-H. Li, H. Fang, J.-G. Xia, and C.-X. Zhu
Microvascular Architecture of Breast Lesions: Evaluation With Contrast-Enhanced Ultrasonographic Micro Flow Imaging
J. Ultrasound Med., June 1, 2008; 27(6): 833 - 842.
[Abstract] [Full Text] [PDF]

eLetters:

Read all eLetters

Why Replace Percutaneous Needle Biopsy of the Breast with Sophisticated New Imaging Techniques?
Stefano Ciatto, MD
Radiology Online, 7 Jun 2007 [Full text]
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2431060041v1
243/1/56    most recent
Right arrow Submit a response
Right arrow View responses
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Chang, R.-F.
Right arrow Articles by Chen, D.-R.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Chang, R.-F.
Right arrow Articles by Chen, D.-R.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE