SS 202 - Breast ultrasound and computer-aided diagnosis (CAD) systems
Purpose: To perform a systematic review and meta-analysis on the detection rate of second look ultrasound(SLUS) in breast lesions detected by Magnet Resonance Imaging(MRI).
Methods and Materials: No ethics committee approval was needed. A systematic literature search was performed (MEDLINE,EMBASE,WEB OF SCIENCE) for articles evaluating SLUS after MRI with follow up or histologic proof as reference standard. Three independent readers selected eligible articles published until April 2016. The quality of studies was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2(QUADAS-2). Effect size with a 95% confidence interval(CI) and heterogeneity of the studies were calculated. The Egger test was performed for assessing the publications bias and a meta-regression was performed to analyze the publication date and the size of the lesions. Subgroup analyses were performed for cancers, mass lesions, non-mass enhancement and foci.
Results: Thirty-three articles were included. The study quality was mostly high. The detection rate of all lesions was very heterogeneous (I2=93.7%; P=0.0001) with a point estimate of 64.2% (95%CI=57.5%-70.4%). The risk of publication bias was significant (P=0.033). Year of publication and mean lesion diameter were not significant (P=0.935 and 0.184, respectively). The detection rate at SLUS of cancers, mass lesions, non-mass enhancement and foci showed a point estimate of 77.5%, 66.5%, 37.5% and 48.1%, respectively.
Conclusion: SLUS is a reasonable examination-tool after finding a previously unknown lesion in breast-MRI, especially in cancerous or mass lesions. Nevertheless negative SLUS does not exclude malignancy and MRI-guided biopsies or follow-up examinations have to be considered for further work-up.
Breast ultrasound: can 3D multiplanar reconstructions aid in the differentiation of benign from malignant lesions?
Purpose: To evaluate the added value of three-dimensional (3D) multiplanar reconstructions (MPRs) to the two-dimensional (2D) B-mode ultrasound (US) for differentiating benign from malignant breast lesions.
Methods and Materials: 99 patients with 101 breast lesions planned to undergo US-guided biopsy were examined with 2D B-mode US. A 3D volume of each lesion was subsequently recorded. 2 readers (R) independently evaluated the 2D image of each lesion according to the breast imaging reporting and data system (BI-RADS) lexicon and assigned a BI-RADS classification. Then, they evaluated the MPRs of each lesion and assigned a new BI-RADS classification, while also recording the presence or absence of retraction in the coronal plane. They were also asked to score their degree of confidence for the BI-RADS classification in a 10-point scale. Receiver-operating characteristics analysis was used to evaluate reader performance and confidence. Logistic regression was used to evaluate the different BI-RADS descriptors in 2D and 3D US.
Results: 53 lesions were malignant and 48 benign. Area under the curve did not differ significantly between 2D and 3D BI-RADS classification for neither of the readers (R1:0.829 vs 0.866, p>0.05, R2:0.847 vs 0.810, p>0.05). Retraction phenomenon showed high specificity (R1:96.55%, R2:89.58%) but low sensitivity (R1:31.82%, R2:42.31%). In multivariate logistic regression significant association with malignancy was shown for margin (p<0.05) in 2D and margin (p<0.001) and shape (p<0.01) in 3D US. A (non-significant) tendency towards higher confidence was shown for 3D US.
Conclusion: 3D MPRs have a similar diagnostic performance as 2D breast US. A positive retraction phenomenon is highly suggestive of malignancy.
Usefulness of computer-aided diagnosis conjunction to breast ultrasound depending on experience of breast imaging
Purpose: To evaluate the usefulness of computer-aided diagnosis (CAD) conjunction to breast ultrasound depending on the experience of breast imaging.
Methods and Materials: Between October 2015 and January 2016, the expert group of two breast radiologists consecutively performed general breast ultrasound for women with screening and diagnostic purposes. When there was suspicious or probable benign lesion, they added the newly developed CAD system (S-detectTM) on them. And, they choose the proper BI-RADS lexicons and categories on general ultrasound, CAD, and general ultrasound with CAD. For the same cases, two first-grade residents without breast imaging experience choose the BI-RADS lexicons and categories on them (general ultrasound, CAD, and general ultrasound with CAD). And then an uninvolved expert radiologist concluded enrolment and assessed the final result correlated with mammography, old ultrasound, and histopathology. Finally we compared the diagnostic performance depending on the experience of breast imaging.
Results: A total of 200 cases were enrolled in this study. These were sensitivities of expert (91.7%), resident (75%), and CAD (75% & 66.7%). These were specificities of CAD (78.2% & 76.1%), expert (76.6%), and resident (71.8%). After combination with CAD, the specificity was improved (76.6% to 80.3%) without change of sensitivity (91.7%) in expert group. After combination with CAD, the sensitivity and specificity were improved in resident group (75% & 71.8% to 83.3% & 77.1%).
Conclusion: CAD is more useful for less experienced radiologists. When combined the CAD to ultrasound, the specificity is improved in any radiologists.
Evaluation of a computer-aided-diagnosis system in breast ultrasound (S-Detect): intrinsic value and effect on junior radiologist’s performance
Purpose: To evaluate diagnosis performance and effect of S-Detect™ on junior radiologist’s interpretation in breast ultrasound. This Computer-Aided Diagnosis (CAD) is based on Breast Imaging Reporting and Data System (BI-RADS).
Methods and Materials: 189 biopsied lesions were included, 111 were malignant and 78 were benign. CAD sensitivity and specificity was calculated. Four junior radiologists read ultrasound images without and with CAD, and their sensitivity (Se), specificity (Sp), and Receiving Operating Characteristic (ROC) analysis were compared. Readers had classified lesions using American College of Radiology (ACR) BI-RADS score, and a binary classification benign/malignant (BM).
Results: CAD sensitivity and specificity were calculated respectively at 82% and 81%. Its use has resulted in an increasing trend with all reader’s sensitivity for either classification used: reader 1: 92% vs. 90% (classification BM and BI-RADS score); reader 2: 78% vs. 77% (classification BM); 91% vs. 88% (BI-RADS score); reader 3: 85% vs. 80% and 98% vs. 97%; and reader 4: 90% vs. 88% (BI-RADS score); and also with the ROC for 3 readers: reader 1: 0.87 vs. 0.86; reader 3: 0.89 vs. 0.87; and reader 4: 0.81 vs. 0.78. These results were not statistically significant (p > 0.01). No significant decrease in specificity was identified after using CAD.
Conclusion: S-Detect™ has good intrinsic performances, but its use did not allow significant modifications for the 4 junior radiologists’ performance; nevertheless, an improving trend could be noted.
Purpose: The aim of this study was to evaluate diagnostic performance of a computer-aided diagnosis (CAD) system for breast ultrasound to distinguish between benign and malignant lesions and to analyse features of lesions interpreted with errors retrospectively.
Methods and Materials: Between October 2015 and August 2016, 338 women (mean age, 48.7± 11.5 years) with 397 lesions were enrolled and underwent breast ultrasound with ultrasound CAD system (S-detectTM). We assessed accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV). In addition, we evaluated causes and patterns of misinterpretation in false positive and negative groups.
Results: Accuracy, sensitivity, specificity, PPV and NPV of breast ultrasound CAD were 77.1 %, 81.6 %, 76.5 %, 32.7 and 96.7 %, respectively. Eight false negative lesions were all oval in shape and parallel in orientation. Among 82 false positive lesions, 31 lesions were benign lesions with suspicious features such as fat necrosis or post-operative change. Second leading cause of misinterpretation in 23 lesions was inappropriate demarcation of lesions due to heterogeneous echogenicity, large size, adjacent parenchyma or posterior acoustic shadowing. 21 lesions with suspicious features with good demarcation and proper description were confirmed as benign histologically. And 7 lesions with good demarcation and descriptions implying benignity showed possible malignancy as a final conclusion.
Conclusion: Breast ultrasound CAD is expected to be helpful in avoiding unnecessary biopsy due to its high NPV. And operators need to know characteristics of lesions prone to misinterpretation and to consider clinical history and findings of other imaging modality.
Purpose: To evaluate quantitative multiparametric ultrasound of the breast for the differentiation of benign and malignant lesions.
Methods and Materials: 118 patients, each with one biopsy-proven, sonographically evident lesion were included in this prospective, IRB-approved study. Each lesion was examined with B-mode ultrasound (US), elastography (Acoustic Radiation Force Impulse-ARFI), Doppler US and Contrast Enhanced US (CEUS). Quantitative indices were recorded for each modality as follows: Shear Wave Velocity (SWV) for ARFI, Pulsatility (PI) and Resistive Index (RI) for Doppler US, and Peak Enhancement (PE), Wash-in Area Under the Curve (WiAUC), Rise Time (RT), mean Transit Time (local) (mTTl), Time To Peak (TTP), Wash-in Rate (WiR), Wash-in Perfusion Index (WiPI), Wash-out AUC (WoAUC), Fall Time (FT) and Wash-out Rate (WoR) for CEUS. Paired and unpaired nonparametric statistics were applied for comparisons as appropriate. Diagnostic accuracy of measurements was compared using Receiver Operating Characteristics (ROC) analysis. Multivariate logistic regression was used to determine independent predictors of malignancy.
Results: 64 lesions were malignant and 54 benign. SWV and RI showed the highest diagnostic performance as measured by the area under the ROC curve (0,871 and 0,805 respectively). At a cut-off value of 3.1 m/s, SWV showed sensitivity of 85.94% and specificity of 85.19%. An RI cut-off of 0.68 revealed sensitivity of 77.78% and specificity of 80.56%. Multivariate logistic regression showed that SWV, RI and mTTl were independent predictors of malignancy.
Conclusion: Multiparametric ultrasound of the breast offers quantitative indices that can aid in the differentiation of benign and malignant breast lesions.
Diagnostic performance of assist strain ratio (ASR) in computing fat-to-lesion ratio (FLR) in ultrasound breast elastography
Purpose: ASR is a new application tool developed to differentiate benign from malignant lesions in breast US elastography. This ASR would potentially eliminate current user dependency in outlining tumour and fat area while computing FLR manually. The objective here is to compare the clinical performance of ASR against manual strain ratio (MSR) in the computation of FLR.
Methods and Materials: 42 breast lesions (24 malignant and 18 benign) scheduled for biopsy were included in this IRB approved and HIIPA complaint study. Skilled physician (RGB) performed the elastography exam and selected a frame for computing FLR. MSR was computed manually while ASR was computed with minimal user input. Hitachi’s HIVISION Ascendus with L75P was used for this study. An average of three measurements was used as a cut-off for differentiating lesions.
Results: Cut-off point were determined to be 2.2 for both MSR and ASR using Youden index. Diagnostic performance of MSR were: sensitivity-96%, specificity-67%, accuracy-83 %, PPV-79%, and NPV-92%. Corresponding performance of ASR were: sensitivity-96%, specificity-83%, accuracy-91%, PPV-89%, and NPV-94%. The AUC for MSR and ASR were 0.86 and 0.95, respectively and average coefficient of variation (COV) were 30% and 43% respectively.
Conclusion: ASR demonstrated excellent diagnostic performance compared to MSR. In addition, COV of ASR is lower than MSR implying reduced intra- and inter-operator dependency. Next, we will use auto-frame-select with ASR to minimise manual dependency in selecting the frame as well.
Purpose: Strain elastography (SE) and shear wave speed imaging provide us with tissue stiffness information, important for breast diagnosis. SE allows us to visualise tissue stiffness that reflects the pathological information. However, it can be associated with operator bias in image formation, scanning technique, the selection of the appropriate frame, and ROI setting for strain ratio measurement. To overcome these problems, we have developed an “Auto Strain Ratio System (ASRS)”, which requires no manual compression whilst scanning, full auto frame selection on freeze and fully automatic ROI target selection. A preliminary prospective clinical study was performed in 2015.
Methods and Materials: 232 breast masses assessed as BI-RADS Category 3 or above by B-mode, were scanned using SE. The new ASRS was compared to conventional manual strain ratio measurement (MSR) by experienced doctors and the diagnostic performance and quality evaluated.
Results: There was a significant correlation between the MSR and ASRS with R=0.79 (p<0.001). The MSR (cut-off=3.8) had a sensitivity of 89%, a specificity of 75%, an accuracy of 78%, a positive predictive value (PPV) of 52%, and a negative predictive value (NPV) of 96%. The ASRS (cut-off=3.9) had a sensitivity of 82%, specificity of 89 %, an accuracy of 87%, a PPV of 69%, and a NPV of 94%. The AUCs were 0.88 on MSR and 0.89 on ASRS.
Conclusion: We have demonstrated that it is possible to quantify SE and control its accuracy. The ASRS is expected to contribute to the standardisation of breast elastography.
Purpose: To determine the sonographic characteristics of core biopsy-proven mass-like focal breast fibrosis (MFBF).
Methods and Materials: IRB approved, retrospective study. Between April 2007 and January 2015, 3051 US-guided breast biopsies with 14G core needle, were performed, 251 of them with a diagnosis of stromal breast fibrosis. We excluded 128 cases where fibrosis was not the primary histologic diagnosis, only MFBF cases were included. Imaging features were tabulated and analysed. Follow-up imaging was reviewed to document lesion stability.
Results: In 121 women (median age: 50 years, range: 25-83) we found 123 cases of MFBF. Lesion size ranged from 4 to 35mm (median: 10mm), non-palpable in 94% of the cases. Eighty-seven (71%) of them developed in dense breast (ACR 4 and 3). Only 7 (6%) were evident on mammography. We identified two distinct sonographic patterns of MFBF. Pattern A (28%): well-circumscribed, hypoechoic, avascular mass. Pattern B (72%): ill defined, irregular, avascular, markedly hypoechogenic, spiculated with shadowing, located intraparenchymatous or under Cooper ligament. Sixty-seven (54%) lesions were reported as BI-RADS 5, 4C or 4B. MRI was performed in 7 patients with negative outcome. One lesion was surgically removed and 4 new biopsies were performed due to discordance, obtaining the same results. Patients remain in follow-up (30 months), without malignancy.
Conclusion: The MFBF is a benign entity with the potential to mimic malignancy. Is important that radiologists know the specific US patterns and if proven on core needle biopsy, it may be taken as a concordant diagnosis.
Clinical usefulness of repeated short-term follow-up imaging in young patients with initial diagnosis of BI-RADS 3 lesions
Purpose: To evaluate the clinical usefulness of repeated short-term follow-up with ultrasound in patients younger than 35 years with a BI-RADS 3 lesion at first ultrasound examination and proven lesion stability at the 6-month follow-up.
Methods and Materials: In this IRB-approved study, 492 women, aged 18-34 years (mean±standard deviation, 28±4.5 years) with first breast ultrasound examination in 2012-2014 were retrospectively evaluated. Inclusion criteria were: at least one BI-RADS3 lesion and (a) biopsy/surgical excision or (b) follow-up of at least 18 months (including a 6-month follow-up). BI-RADS category assigned during follow-up and pathologic findings in cases undergoing biopsy/surgical excision were collected. Recommended biopsy rates (RBR) after 6 and 18 months and positive predictive value (PPV) for biopsy after recommendation due to interval changes (PPVbio) were calculated.
Results: In 97 patients, 151 BI-RADS3 lesions were identified. Biopsy/surgical excision was performed after initial assessment in 25/151 (16.5%) lesions. At 6-month follow-up, assessment category was changed to BI-RADS 1 or 2 in 23/126 (15.3%) and to BI-RADS 4 in 9/126 lesions (7.1%) due to interval growth. Pathological diagnosis of these lesions was fibroadenoma in 5 cases and benign phyllodes tumour in 4 cases (RBR 7%, PPVbio 44.4%). At 18-month follow-up one lesion was classified BI-RADS 4 due to interval growth and pathological diagnosis was fibroadenoma (RBR 1.1%, PPVbio 0%).
Conclusion: Follow-up imaging performed after 18 months from a first BI-RADS3 diagnosis does not affect clinical treatment and 6-month follow-up may be sufficient to assess the stability of probably benign lesions and to discern those which entail further investigation.