| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 All authors: Breast Imaging, Department of Radiology, University of Washington and the Seattle Cancer Care Alliance, 825 Eastlake Ave. East, G3-200 PO Box 19023, Seattle, WA 98109-1023.
Received February 16, 2005;
accepted after revision May 13, 2005.
Presented at the 2003 annual meeting of the Radiological Society of North
America, Chicago, IL.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Thirty-three consecutive lesions seen only on MRI (nine malignant, 24 benign) were evaluated with and without the automated software system. Automated analyses of kinetic enhancement for each lesion were recorded at 50%, 80%, and 100% enhancement thresholds. Computer-assisted analyses included presence or absence of "significant" enhancement and classification of enhancement patterns into percent volumes of washout, plateau, and persistent enhancement. Fisher's exact tests were performed to compare the likelihood of malignancy based on the presence of software-defined significant enhancement at the three thresholds. Enhancement profiles of malignant versus benign lesions were compared using the Student's t test.
RESULTS. All malignant lesions showed significant enhancement at all thresholds. Compared with the unassisted interpretations, the computer-assisted analyses yielded false-positive rates that were reduced by 25% at a 50% threshold (not significant [NS]), 33% at an 80% threshold (p = 0.05), and 50% at a 100% threshold for enhancement (p < 0.01). There were no significant differences between enhancement profiles of benign and malignant lesions, with all lesions showing a wide range of washout, plateau, and persistent patterns of enhancement.
CONCLUSION. New automated software applied to interpret breast MR examinations accurately showed significant enhancement in all the malignant lesions while depicting 12 of 24 benign lesions as showing insignificant enhancement. If these results are validated by a larger study, the number of unnecessary biopsies of MR lesions could be reduced without a concomitant decrease in cancer detection.
Keywords: BI-RADS • breast cancer • computer-aided detection • mammography • MRI
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Nevertheless, breast MRI has not achieved widespread use in clinical practice. A number of distinct barriers prohibit more extensive use of MRI as a complement to mammography: high cost; variable specificity [16, 17]; and extensive time required by both the technologist and the radiologist in image acquisition, processing, and interpretation. Computer-aided image management and analysis have the potential to impact each of these obstacles, providing tools to improve the diagnostic accuracy (particularly through improved specificity), consistency, and efficiency of breast MR image interpretation.
A new commercially available method of image processing and evaluation has been developed for breast MRI. We conducted this study to compare the accuracy of breast MRI interpretations with and without this automated software program.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
All study examinations were performed on a 1.5-T scanner (LX, GE Healthcare) using a dedicated bilateral breast coil (GE 1.5T 4 Channel Breast Array, MRI Devices). The MRI scan protocol parameters include both unenhanced and contrast-enhanced sagittal T1-weighted 3D fast spoiled gradient-recalled (FSPGR) sequences: TR/TE, 6.7/4.2; flip angle, 35°; and matrix, 256 x 192. The field of view was restricted to 20-22 cm over the study breast depending on patient size, and slices were 3 mm or less in thickness. T1 3D FSPGR images were acquired before and immediately after bolus injection of contrast material. Immediate contrast-enhanced images were acquired within 4 minutes and delayed contrast-enhanced images were acquired within 8 minutes of contrast injection. Scanning protocols adhered to the guidelines established by the International Breast MRI Consortium and by the ACR Imaging Network (ACRIN) MRI trials.
All MRI examinations were prospectively interpreted by one or more radiologists specializing in breast imaging. All suspicious MRI enhancing lesions were described in terms of shape, margin, distribution, and internal enhancement pattern. MRI interpretation criteria, assessments, and recommendations were based on the breast MRI lexicon as set forth in the ACR BI-RADS [18]. Enhancement kinetics were prospectively assessed subjectively by visual inspection and classified as persistent, plateau, or washout of contrast material. For all BI-RADS MRI category 4 (suspicious abnormality) and category 5 (highly suggestive of malignancy) lesions, a targeted second-look sonography examination was recommended and performed. Lesions visible on sonography were biopsied using sonographic guidance and were not included in this study.
Automated Software Program
All MRI examinations were retrospectively processed by CADstream
(Confirma). This software is designed to automate image processing and
analysis functions that are typically performed manually by the MR
technologist and radiologist. All processed data are saved and presented with
the original images to the radiologist for interpretation. The software
program produces color overlays on all image slices in a series to identify
areas of significant enhancement. In addition, it provides an automated
interactive display of kinetic enhancement curves and details about all
regions of significant enhancement.
For breast MR examinations, the program incorporates three MR series into its calculations: one unenhanced T1-weighted series, one immediate contrast-enhanced T1-weighted series, and one delayed contrast-enhanced T1-weighted series. The specific contrast-enhanced series used at a given practice site is determined by the radiologist, depending on the site's scanning protocol. For most sites, it is suggested that the immediate contrast-enhanced series be centered over roughly 2 minutes (images obtained within 4 minutes) and the delayed contrast-enhanced series be centered over 4-7 minutes (images obtained within 8 minutes).
To classify areas on the MR examination as having or not having "significant" enhancement, pixel values at the unenhanced and first contrast-enhanced series are compared. If the pixel value increases by an established threshold, the pixel will be shown in color on the monitor. If the pixel value fails to increase by the established threshold, no color enhancement will be made. In addition, the program assigns a specific color to each pixel that meets the threshold for significant enhancement. The color assigned depends on the change in the pixel values between the initial contrast-enhanced series and the delayed contrast-enhanced series. If the pixel value on the delayed series decreases by more than 10% of its immediate contrast-enhanced value, that pixel is color-coded red on the monitor, indicating washout of contrast material. If the pixel value increases by more than 10%, it is color-coded blue on the monitor, indicating persistent enhancement. If the pixel value does not change in either direction by more than 10%, it is color-coded green for plateau enhancement (Fig. 1). The colors generated by the program can be changed according to site preference. The end result is a color overlay on each MRI slice indicating regions of significant enhancement and providing details about enhancement type and extent. Finally, the radiologist may select a specific area of significant enhancement and the program will automatically generate a synopsis of the full volume of that lesion, including the percentage of the lesion that shows washout, plateau, and persistent enhancement (Figs. 2A, 2B, 2C, 2D, and 2E). Once a lesion is selected by the radiologist, this information generated by the program is fully automated.
|
|
|
|
|
|
Data Analysis
Fisher's exact tests were performed to compare the likelihood of malignancy
based on the presence of significant enhancement at each of the three
enhancement thresholds. Enhancement profiles of malignant versus benign
lesions were compared using the Student's t test.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Accuracy of Computer-Defined Significant Enhancement
All malignant lesions showed significant enhancement using the program at
all thresholds, producing a sensitivity of 100%. Benign lesions varied in
significant enhancement across the three thresholds.
Figure 3 shows the presence of
significant enhancement for the 33 lesions at the three different thresholds
compared with the original radiologist interpretation without computer
assistance. The number of false-positive examinations decreased as the
enhancement threshold increased from 50% to 80% and from 80% to 100%.
False-positive rates for the computer-aided assessment compared with the
original radiologist assessment were reduced by 25% at a 50% threshold (NS),
33% at an 80% threshold (p = 0.05), and 50% at a 100% threshold
(p < 0.01) for enhancement.
|
No significant differences were noted between the delayed contrast-enhanced kinetic profiles of benign and malignant lesions, with all lesions showing a wide range of washout, persistent, and plateau patterns of enhancement. There was significant overlap in the appearance of benign and malignant lesions, with some malignant lesions exhibiting marked persistent enhancement and some benign lesions showing a significant degree of washout of contrast material (Fig. 4).
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The ACR BI-RADS lexicon for breast MRI recommends reporting both the initial and the delayed enhancement characteristics on all lesions. We found the information from the initial portion of the time-signal intensity curve provided by the CADstream software was most predictive of malignancy. Lesions that failed to reach an initial peak enhancement threshold on the first contrast-enhanced series were significantly more likely to be benign. This is consistent with results reported by Szabo et al. [19], who evaluated 109 pathology-verified breast lesions on MRI. In that study of both kinetic and morphologic diagnostic predictors, the time to peak enhancement (initial phase of enhancement) was a significant independent predictor of malignancy. Neither our study nor that of Szabo et al. found information from the delayed portion of the curve (washout, plateau, or persistent enhancement) to be helpful in distinguishing benign from malignant lesions. Similarly, Stomper et al. [20] and Orel et al. [21] found significant overlap between time-intensity curves of benign versus malignant lesions.
Others have reported that delayed contrast activity is predictive of histology [22, 23]. Kuhl et al. [22] reported on 266 pathology-proven lesions and found that the shape of the time-signal intensity curve was an important criterion in differentiating benign from malignant lesions. A pattern of washout of contrast material was significantly more likely to be seen in malignant lesions, and a pattern of persistent enhancement was significantly more likely to be associated with benign lesions. However, there was overlap in the patterns of benign and malignant lesions and Kuhl et al. cautioned against using curve shape alone to assess the likelihood of malignancy.
There are several explanations for the differences between our findings and those of Kuhl et al. [22]. At this time, the methods of image acquisition and timing of delayed contrast-enhanced series vary widely across sites. The software method used in this study uses two contrast-enhanced series, whereas prior reported methods of curve assessment can use as many as nine contrast-enhanced series. In addition, this method provides detailed information about each pixel in a given tumor, whereas prior techniques require the radiologist or technologist to place a region of interest (ROI) on the most suspicious area of the lesion after visual inspection of the images. The latter method produces one curve, rather than multiple curves with a summary assessment of all pixels in the lesion.
As this method of image analysis and others continue to be developed for breast MRI, numerous challenges will need to be addressed. Most importantly, there is no clear consensus about how best to acquire breast MR images. Currently, there are numerous methods of acquiring images for an acceptable breast MR examination, including unilateral versus bilateral examinations, examinations with high spatial resolution versus examinations with high temporal resolution, and countless post-processing algorithms. Computer-assisted diagnosis (CAD) programs designed for one method of examination may not apply to others. Thus, it will be important to evaluate emerging CAD programs across a broad range of image acquisitions.
It is interesting that most of the lesions in our study had heterogeneous patterns of enhancement. In other words, although the ACR BI-RADS lexicon recommends a pattern of delayed contrast enhancement be assigned to each lesion (washout, plateau, or persistent), it is the rare lesion that shows only one pattern of enhancement. Our study shows that when each pixel is assessed rather than an average assessment from a larger ROI of the lesion, multiple different curves are generated for each lesion. The ACR lexicon suggests that the most suspicious pattern of enhancement be reported when more than one curve is identified; however, there was at least a small percentage of washout shown in almost all lesions in our study. It may be useful to evaluate the relevance of heterogeneous versus homogeneous patterns of enhancement of lesions in predicting histology.
There are limitations to this study. This was a single-site, retrospective study of consecutive suspicious MR lesions recommended for biopsy, and all malignant lesions were invasive. Whether our findings can be generalized to other methods of image acquisition is not clear. However, our method of acquisition is in keeping with the standards for image acquisition established by the International Breast MRI Consortium [24] and by the two major ACRIN studies on breast MRI (ACRIN 6657 and ACRIN 6667). It is important to recognize the specific patient population evaluated in this study. Because the cancers in this study were all invasive carcinomas, these results may not apply to cases of ductal carcinoma in situ. In addition, the software program is limited in evaluating lesions surrounded by diffuse background enhancement. Specifically, five of 21 lesions could not be assessed independently from diffuse enhancement in the breast. This limitation will need to be addressed in future programs.
These preliminary results are promising. In addition to improving specificity, automated software programs should contribute to the field of breast MRI by providing more objective and detailed information regarding MR examinations and lesions. In this way, software programs may be able to help direct improvements in scan protocols more rapidly than reviewer studies or other methods of subjective assessments of image quality. These programs may decrease heterogeneity of interpretations across radiologists of varying levels of experience in breast MR interpretations. Finally, as MR technology continues to rapidly advance and to provide more image data on each patient scanned, these types of programs have the potential to decrease the amount of time required for image processing and interpretation. Additional studies are being performed to evaluate more clearly the ability of software programs to address challenges that delay more widespread dissemination of breast MRI in appropriate clinical populations.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  T. C. Williams, W. B. DeMartini, S. C. Partridge, S. Peacock, and C. D. Lehman Breast MR Imaging: Computer-aided Evaluation Program for Discriminating Benign from Malignant Lesions Radiology, July 1, 2007; 244(1): 94 - 103. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
