Role of sonoelastography in non-palpable breast lesions

Eur Radiol (2008) 18: 2381–2389 DOI 10.1007/s00330-008-1032-8 BREAST Gianfranco Scaperrotta Claudio Ferranti Claudia Costa Luigi Mariani Monica Marc...
Author: Jessie Bradley
4 downloads 2 Views 236KB Size
Eur Radiol (2008) 18: 2381–2389 DOI 10.1007/s00330-008-1032-8

BREAST

Gianfranco Scaperrotta Claudio Ferranti Claudia Costa Luigi Mariani Monica Marchesini Laura Suman Cristina Folini Silvana Bergonzi

Role of sonoelastography in non-palpable breast lesions

Received: 28 September 2007 Revised: 7 March 2008 Accepted: 30 March 2008 Published online: 4 June 2008 # European Society of Radiology 2008

G. Scaperrotta (*) Fondazione IRCCS Istituto Nazionale Tumori, Via Venezian 1, 20133 Milan, Italy e-mail: gianfranco. [email protected] Tel.: +39-223-902519 Fax: +39-223-902524

G. Scaperrotta . C. Ferranti . C. Costa . M. Marchesini . L. Suman . C. Folini . S. Bergonzi Department of Breast Imaging, Fondazione IRCCS Istituto Nazionale Tumori, Milan, Italy L. Mariani Unit of Medical Statistics and Biometry, Fondazione IRCCS Istituto Nazionale Tumori, Milan, Italy

Abstract The purpose of this study was to evaluate the diagnostic utility of sonoelastography in differentiating benign from malignant non-palpable breast lesions. A total of 293 BIRADS 3–5 (Breast Imaging Reporting And Data System) impalpable breast lesions in 278 women was evaluated with B-mode ultrasound (US) and subsequently with sonoelastography (SE) before performing US-guided biopsy. Among the 293 lesions (size up to 2 cm), 110 (37.5%) were histologically malignant and 183 (62.5%) benign. Lesions that were malignant or showed atypical ductal hyperplasia

Introduction Breast ultrasonography (US) has become an invaluable tool for detection and imaging-guided biopsy of impalpable lesions, particularly in women with dense breasts. Technological improvements in transducer resolution and signal processing have enabled sensitivity of 85–95% and specificity of 76–80% to be achieved [1, 2]. US, however, is still strongly operator-dependant, and a correct diagnosis may be sometimes difficult because of the overlapping between the features of malignant and

were referred for surgical excision, as well as 32 benign lesions showing discordance between US/SE results and histology. All other benign lesions had US follow-up at 6/12 months, showing stability. Overall performance of SE was lower than US, with sensitivity and specificity of 80% and 80.9%, respectively, for SE as compared with 95.4% and 87.4% for US. Statistical analysis showed no improvement in the joint use of SE and US over the use of US alone, whose performance, however, was very high in our study. SE is a simple, fast and non-invasive diagnostic method that may be a useful aid to US for less experienced radiologists in the assessment of solid non-palpable breast lesions, especially BI-RADS 3, where specificity was higher (88.7%). Keywords Sonoelastography . Breast ultrasound . Non-palpable breast lesions

benign breast lesions, although they have been described [3–5] and categorized [6]. Consequently, the diagnostic confirmation may often require image-guided biopsy procedures. Recently, a new ultrasound approach has been introduced, sonoelastography (SE), based on the principle that malignant tissue is harder than benign tissue. Elastography was introduced in 1991 [7] and started to be used in a clinical setting in 1997 [8]. Various methods have been tested before real-time free-hand SE was available, and Ueno and co-authors [9] have proposed

2382

simple interpretative criteria and a scoring system that classifies the lesions in the same way used by BI-RADS for classification of B-mode US images [6]. In preliminary papers authors have reported cases series including lesions up to 3 cm in diameter and sometimes palpable, with SE showing lower sensitivity, but greater specificity than US [9–11]. The reported sensitivity and specificity of SE range between 79–100% and 79–95%, respectively [4], and the exam is considered a promising opportunity because it is sufficiently easy to perform after an adequate training and takes only about 5 min after US assessment. We adopted a revised SE score system, based on a multicenter Italian Team of Study [12], which was used to evaluate 293 non-palpable breast lesions up to 2 cm in diameter before performing US-guided biopsy. The purpose was to determine the performance of the exam and estimate which role it may play in daily workup of nonpalpable lesions.

On the scheduled day, for each patient the lesion/lesions was/were correctly identified with conventional B-mode US on the basis of previous mammographic and US studies, then a directed SE was performed on the region/ regions of interest (ROI). Finally, the lesions were submitted to US-guided 18- or 14-gauge core-needle biopsy (CNB) with an automatic short-throw (1.2 cm) biopsy gun (Pro-Mag). Only in 4 out of 293 cases, fineneedle aspiration citology (FNAC) was performed instead of CNB. Lesions that were malignant or showed atypical hyperplasia at CNB were referred for excision, while among lesions with benign CNB results 32 patients were referred for confirmatory open surgical biopsy, due to persistent risk of underestimation by discordance between US/SE results and histological diagnosis. The definitive histologic results were used as reference. Lesions with benign CNB results were submitted to US follow-up at 6 and 12 months. No significative changes concerning morphology and size of the lesions were found.

Materials and methods Equipment Patients A prospective study was performed from January 2005 to May 2006 at the Breast Imaging Unit of the National Cancer Institute of Milan to assess the diagnostic capability of sonoelastography in non-palpable breast lesions prior to US-guided needle biopsy, with biopsy results as reference standard. Lesions that were clearly cystic or benign at B-mode US were excluded because we intended to enroll only lesions that could be liable to investigation (BI-RADS 3) or suspicious (BI-RADS 4–5) according to the Breast Imaging Reporting and Data System lexicon of the American College of Radiology [6]. Our study enrolled 278 women, aged from 29 to 82 years (mean of 53) with a total of 293 focal lesions, ranging from 3 mm to 20 mm in diameter. This number represents the set of eligible cases, after the exclusion of 17 cases because of unsatisfactory image quality (unacceptable elastographic image parameters). Lesions measuring more than 2 cm were also not included in the analysis. Among the lesions detected at conventional B-mode US, 165 were were classified as BI-RADS 3 being represented by developing or enlarging benign appearing masses. In the subset of 128 suspicious cases, classified as BI-RADS 4 or 5, 116 were represented by masses and 12 by focal architectural distortions. Most of the patients submitted to US-guided biopsy had previous mammography and/or ultrasound at our unit, but several women were referred to us on the basis of studies performed elsewhere.

Pre-biopsy conventional US was performed using a digital electronic US unit LOGOS HiVision ESAOTE/HITACHI, with a linear array broad-band transducer with a frequency range of 6–14 MHz. Sonoelastography was performed with the same equipment, employing a dedicated software with an algorithm that in a very short time processes the radio-frequency impulses coming from the lesion and displays in real time and in color scale the degree of tissue strain in the ROI. Elastographic method After B-mode US detection of the lesion of interest, the patient remains in the supine position, and a stabilizer device is mounted on the probe to hold an homogeneous pressure on a wider area of the skin’s surface by minimizing lateral movements of the probe. Then the dual elastographic program starts, with the US monitor showing in real time the B-mode US image of the lesion on the right side and the same image with colorcoded elasticity features superimposed on the left side. The SE exam usually lasts on average 5 min, and motion images are obtained by applying a light constant pressure with the probe in contact with the skin perpendicular to the chest wall. In order to obtain correct elastographic images, attention must be paid to the definition of the ROI, which has to be sufficiently wide to include enough breast tissue surrounding the lesion so that data about the average strain of the tissue inside the region are available. The ROI usually must extend from the

2383

subcutaneous fat at the top to the anterior profile of the pectoral muscle at the bottom, with lateral borders set more than 5 mm from the lesion’s boundary. The exam is correctly performed checking the 1–5 LED scale that appears laterally on the right of the elastographic image that is indicative of proportionality between pressure and tissue strain. The color-coded image must be consistently overlapped with the B-mode US image, with a smooth appearance and without color flashes. The elasticity images are obtained according to a 256color scale ranging from red, indicative of the softest tissues that show the greatest strain, to blue for the hardest components that don’t exhibit any strain, with green corresponding to the average strain observed in the ROI. To classify elastographic images, the 5-score system proposed by Ueno and co-workers [9] was considered, because it can be easily correlated to the 5-score BI-RADS classification, thus allowing a practical management of the lesions. However, a slight adjustment of Ueno scoring descriptors was undertaken according to the panel assessment of an Italian Multi-Centric Team of Study for Sonoelastography Evaluation in which we actively participated [12]. In the Ueno classification [9], score 1 indicated strain for the whole hypoechoic lesion, which appeared shaded in green. Score 2 indicated strain in most of the hypoechoic lesion, with a mosaic pattern of green and blue. Score 3 indicated strain (green) at the periphery of the lesion with a stiffness blue center. A score 4 indicated stiffness in the entire hypoechoic lesion, which was totally blue. Finally, score 5 indicated no strain in the entire hypoechoic lesions and in the surrounding area, both appearing blue. The Italian Team of Study classification [12] that we adopted differs from the above-mentioned one mainly for the score 1 lesions, which exhibit a typical three-layer feature (blue-green-red from the surface to the bottom) usually indicative of cystic lesions. This is a feature that Ueno and co-workers described in their preliminary report, but it is not referred to in their latest extensive paper [9]. In our classification, score 2 is a benign-like lesion almost entirely green with random blue points. A score 3 is a lesion predominantly green showing some blue spots, consistent with benignity. Score 4 is an almost entirely blue lesion with minimal green points at the periphery, suspect for malignancy. Score 5 is the same as in the Ueno classification, with an entirely blue lesion surrounded by a blue halo, consistent with malignancy (Fig. 1). Statistical methods Lesions were classified at US B-mode according to BIRADS (Breast Imaging Reporting and Data System). The

CHROMATIC CODE

ELASTOSONOGRAPHIC SCORE SCORE 1: Presence of chromatic tri-stratification (blue /green / red)

ITALIAN TEAM OF STUDY Prevalently in the liquid forms

SCORE 2: Prevalence of green, with in case some blue point, inconstant seat SCORE 3: Prevalently green, but with some blue spot.

PREVALENTLY ELASTIC: prevalently in the benign forms

SCORE 4: Almost completely blue, with in case some green point, most of all in periphery SCORE 5: Completely blue, also with a blue peripheral glow around the nodule

PREVALENTLY RIGID: prevalently in the malignant forms

Fig. 1 Sonoelastographic classification by the Italian Multi-Center Team of Study

correlation between US and SE scores was assessed by calculating the Kendall’s tau-correlation coefficient: a “0” correlation coefficient denotes the lack of association between two scores, whereas a value of “1” would indicate perfect association. Considering each score level as a possible threshold for test positiveness, true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN) were computed, and their frequency was used to obtain the estimates of sensitivity (TP/[TP+FN]), specificity (TN/ [TN+FP]), positive predictive value (PPV=TP/[TN+FN]) and negative predictive value (NPV=TN/[TN+FN]). Histology was used as reference for defining the lesion nature (benign, malignant). The overall diagnostic performance of the two methods under investigation, which is independent of the choice of a specific test threshold, was assessed by means of the Area under the ROC Curve (AUC-ROC), calculated nonparametrically as proposed by Hanley and McNeil [14]. An AUC-ROC of “1” indicates perfect discrimination between malignant and benign lesions, a value of “0.5” indicates the lack of discrimination, and values in between indicate partial discrimination that may vary from poor to strong. The comparison between the curves for US and ES, being derived from the same lesions, was performed as described in [15], whereas the remaining comparisons between independent sub-groups were carried out as described in [14]. Subgroup analyses were carried out in specific subsets, such as SE scores in the case of different US categories, or for lesions measuring up to 10 mm vs. 11–20 mm. Furthermore, the AUC-ROC estimate for the joint use of US and ES was estimated by fitting a multiple logistic regression model in which the status of the lesion (benign or malignant) was entered as the response variable,

2384

whereas the US and SE scores were entered as the predictors.

Results At histology 110 lesions resulted in being malignant (37.5%) and 183 benign (62.5%) (Table 1). Lesion size was up to 5 mm in 66 cases (22.5%), 6– 10 mm in 133 cases (45.4%), 11–15 mm in 65 (22.2%) and 16–20 mm in 29 cases (9.9%). Lesion size was not significantly different between the two subsets, with 122/183 (66%) benign lesions measuring up to 10 mm as compared with 77/110 (70%) malignant lesions. Patients with malignancy were on average older than women with benign diagnosis (mean age 60.1 years vs. 48.7). Among benign cases, apart from specific diagnoses like fibroadenoma or papilloma or cyst, the other cases were reported as fibrosclerosis, chronic mastitis and ANDI (Aberrations of Normal Development and Involution of breast parenchyma, according to the nomenclature by Hughes) [13]. The remaining five cases classified as “other” included two intramammary lymph nodes, one hemangioma and two cases reported as “lipomatosis.” Three lesions yielded atypical results at CNB. In one case subsequent excision with vacuum-assisted breast biopsy didn’t confirm atypia or malignancy. The same occurred for another case that was submitted to surgical

Table 1 Histologic diagnoses in 293 non-palpable breast lesions Pathologic diagnosis

No. of cases

Percent

Malignant lesions -Ductal invasive carcinoma -Lobular invasive carcinoma -Mucinous carcinoma -Tubular carcinoma -DCIS -LCIS Benign lesions -Fibroadenoma -Papilloma -ANDI -Atypical hyperplasia -Fibrosclerosis -Chronic mastitis -Cyst -Other

110 78 16 4 3 8 1 183 58 9 54 2 36 14 5 5

37.5 26.6 5.5 1.4 1.0 2.4 0.3 62.5 19.8 3.0 18.6 0.7 12.3 4.8 1.6 1.6

biopsy. In the third case surgery revealed multifocal disease, so in our series we have considered this case as malignant despite the underestimation at CNB. Among the 293 non-palpable breast lesions, the relationship between US and SE scores is shown in Table 2. As suggested by the relatively high number of cases along the diagonal of the table, the scores of the two methods tended to be positively correlated. This was confirmed by an estimated Kendall’s tau-correlation coefficient of 0.60 (standard error, 0.034). The distribution of US and SE scores separately for benign and malignant lesions is shown in Table 3, together with sensitivity and specificity. For US, high sensitivity and specificity were achieved with a threshold of 3, which is considering negative for malignancy a BI-RADS score 3 and as positive a score 4 or 5. Correspondent PPV and NPV (with 95% confidence limits) were 0.820 (0.743–0.883) and 0.970 (0.931–0.990). The AUC-ROC estimate was 0.926, indicating a strong and statistically significant (P

Suggest Documents