Should We Ignore, Follow, or Biopsy? Impact of Artificial Intelligence Decision Support on Breast Ultrasound Lesion Assessment.

OBJECTIVE. The objective of this study was to assess the impact of artificial intelligence (AI)-based decision support (DS) on breast ultrasound (US) lesion assessment. MATERIALS AND METHODS. A multicenter retrospective review of 900 breast lesions (470/900 [52.2%] benign; 430/900 [47.8%] malignant) on US by 15 physicians (11 radiologists, two surgeons, two obstetrician/gynecologists). An AI system (Koios DS for Breast, Koios Medical) evaluated images and assigned them to one of four categories: benign, probably benign, suspicious, and probably malignant. Each reader reviewed cases twice: 750 cases with US only or with US plus DS; 4 weeks later, cases were reviewed in the opposite format. One hundred fifty additional cases were presented identically in each session. DS and reader sensitivity, specificity, and positive likelihood ratios (PLRs) were calculated as well as reader AUCs with and without DS. The Kendall τ-b correlation coefficient was used to assess intraand interreader variability. RESULTS. Mean reader AUC for cases reviewed with US only was 0.83 (95% CI, 0.78-0.89); for cases reviewed with US plus DS, mean AUC was 0.87 (95% CI, 0.84-0.90). PLR for the DS system was 1.98 (95% CI, 1.78-2.18) and was higher than the PLR for all readers but one. Fourteen readers had better AUC with US plus DS than with US only. Mean Kendall τ-b for US-only interreader variability was 0.54 (95% CI, 0.53-0.55); for US plus DS, it was 0.68 (95% CI, 0.67-0.69). Intrareader variability improved with DS; class switching (defined as crossing from BI-RADS category 3 to BI-RADS category 4A or above) occurred in 13.6% of cases with US only versus 10.8% of cases with US plus DS (p = 0.04). CONCLUSION. AI-based DS improves accuracy of sonographic breast lesion assessment while reducing inter- and intraobserver variability.

Fully Automated Convolutional Neural Network Method for Quantification of Breast MRI Fibroglandular Tissue and Background Parenchymal Enhancement.

The aim of this study is to develop a fully automated convolutional neural network (CNN) method for quantification of breast MRI fibroglandular tissue (FGT) and background parenchymal enhancement (BPE). An institutional review board-approved retrospective study evaluated 1114 breast volumes in 137 patients using T1 precontrast, T1 postcontrast, and T1 subtraction images. First, using our previously published method of quantification, we manually segmented and calculated the amount of FGT and BPE to establish ground truth parameters. Then, a novel 3D CNN modified from the standard 2D U-Net architecture was developed and implemented for voxel-wise prediction whole breast and FGT margins. In the collapsing arm of the network, a series of 3D convolutional filters of size 3 × 3 × 3 are applied for standard CNN hierarchical feature extraction. To reduce feature map dimensionality, a 3 × 3 × 3 convolutional filter with stride 2 in all directions is applied; a total of 4 such operations are used. In the expanding arm of the network, a series of convolutional transpose filters of size 3 × 3 × 3 are used to up-sample each intermediate layer. To synthesize features at multiple resolutions, connections are introduced between the collapsing and expanding arms of the network. L2 regularization was implemented to prevent over-fitting. Cases were separated into training (80%) and test sets (20%). Fivefold cross-validation was performed. Software code was written in Python using the TensorFlow module on a Linux workstation with NVIDIA GTX Titan X GPU. In the test set, the fully automated CNN method for quantifying the amount of FGT yielded accuracy of 0.813 (cross-validation Dice score coefficient) and Pearson correlation of 0.975. For quantifying the amount of BPE, the CNN method yielded accuracy of 0.829 and Pearson correlation of 0.955. Our CNN network was able to quantify FGT and BPE within an average of 0.42 s per MRI case. A fully automated CNN method can be utilized to quantify MRI FGT and BPE. Larger dataset will likely improve our model.

Prior to Initiation of Chemotherapy, Can We Predict Breast Tumor Response? Deep Learning Convolutional Neural Networks Approach Using a Breast MRI Tumor Dataset.

We hypothesize that convolutional neural networks (CNN) can be used to predict neoadjuvant chemotherapy (NAC) response using a breast MRI tumor dataset prior to initiation of chemotherapy. An institutional review board-approved retrospective review of our database from January 2009 to June 2016 identified 141 locally advanced breast cancer patients who (1) underwent breast MRI prior to the initiation of NAC, (2) successfully completed adriamycin/taxane-based NAC, and (3) underwent surgical resection with available final surgical pathology data. Patients were classified into three groups based on their NAC response confirmed on final surgical pathology: complete (group 1), partial (group 2), and no response/progression (group 3). A total of 3107 volumetric slices of 141 tumors were evaluated. Breast tumor was identified on first T1 postcontrast dynamic images and underwent 3D segmentation. CNN consisted of ten convolutional layers, four max-pooling layers, and dropout of 50% after a fully connected layer. Dropout, augmentation, and L2 regularization were implemented to prevent overfitting of data. Non-linear functions were modeled by a rectified linear unit (ReLU). Batch normalization was used between the convolutional and ReLU layers to limit drift of layer activations during training. A three-class neoadjuvant prediction model was evaluated (group 1, group 2, or group 3). The CNN achieved an overall accuracy of 88% in three-class prediction of neoadjuvant treatment response. Three-class prediction discriminating one group from the other two was analyzed. Group 1 had a specificity of 95.1% ± 3.1%, sensitivity of 73.9% ± 4.5%, and accuracy of 87.7% ± 0.6%. Group 2 (partial response) had a specificity of 91.6% ± 1.3%, sensitivity of 82.4% ± 2.7%, and accuracy of 87.7% ± 0.6%. Group 3 (no response/progression) had a specificity of 93.4% ± 2.9%, sensitivity of 76.8% ± 5.7%, and accuracy of 87.8% ± 0.6%. It is feasible for current deep CNN architectures to be trained to predict NAC treatment response using a breast MRI dataset obtained prior to initiation of chemotherapy. Larger dataset will likely improve our prediction model.

Evaluation of association between degree of background parenchymal enhancement on MRI and breast cancer subtype.

PURPOSE: Evaluate possible association between BPE and breast cancer tumor type/prognostic markers. METHODS: IRB approved retrospective study from 1/2010-1/2014 identified 328 patients who had breast MRI and available clinical/pathology data. BPE was categorized according to BI-RADS. The association between BPE and breast cancer molecular subtype/prognostic factors was evaluated. RESULTS: No significant association was present between high BPE and the following: HER2+ tumors, basal tumors, tumors with axillary nodal disease, high nuclear grade tumors, high Ki-67 index tumors or larger tumors. CONCLUSION: Patients with high BPE may be at increased risk for breast cancer but not necessarily for those cancer subtypes with a poor prognosis.

Axillary Lymph Node Evaluation Utilizing Convolutional Neural Networks Using MRI Dataset.

The aim of this study is to evaluate the role of convolutional neural network (CNN) in predicting axillary lymph node metastasis, using a breast MRI dataset. An institutional review board (IRB)-approved retrospective review of our database from 1/2013 to 6/2016 identified 275 axillary lymph nodes for this study. Biopsy-proven 133 metastatic axillary lymph nodes and 142 negative control lymph nodes were identified based on benign biopsies (100) and from healthy MRI screening patients (42) with at least 3 years of negative follow-up. For each breast MRI, axillary lymph node was identified on first T1 post contrast dynamic images and underwent 3D segmentation using an open source software platform 3D Slicer. A 32 × 32 patch was then extracted from the center slice of the segmented tumor data. A CNN was designed for lymph node prediction based on each of these cropped images. The CNN consisted of seven convolutional layers and max-pooling layers with 50% dropout applied in the linear layer. In addition, data augmentation and L2 regularization were performed to limit overfitting. Training was implemented using the Adam optimizer, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments. Code for this study was written in Python using the TensorFlow module (1.0.0). Experiments and CNN training were done on a Linux workstation with NVIDIA GTX 1070 Pascal GPU. Two class axillary lymph node metastasis prediction models were evaluated. For each lymph node, a final softmax score threshold of 0.5 was used for classification. Based on this, CNN achieved a mean five-fold cross-validation accuracy of 84.3%. It is feasible for current deep CNN architectures to be trained to predict likelihood of axillary lymph node metastasis. Larger dataset will likely improve our prediction model and can potentially be a non-invasive alternative to core needle biopsy and even sentinel lymph node evaluation.

The unusual suspects: A review of unusual benign and malignant male breast imaging cases.

Male breast disease is uncommon. Men presenting with breast symptoms may represent unique diagnostic challenges for the radiologist, particularly if imaging findings are not classic for gynecomastia or carcinoma. In this paper we review 10 unusual male breast cases, 5 benign and 5 malignant, including the radiologic findings, differential diagnosis, pathology and management.

Does breast MRI background parenchymal enhancement indicate metabolic activity? Qualitative and 3D quantitative computer imaging analysis.

PURPOSE: To investigate whether the degree of breast magnetic resonance imaging (MRI) background parenchymal enhancement (BPE) is associated with the amount of breast metabolic activity measured by breast parenchymal uptake (BPU) of 18F-FDG on positron emission tomography / computed tomography (PET/CT). MATERIALS AND METHODS: An Institutional Review Board (IRB)-approved retrospective study was performed. Of 327 patients who underwent preoperative breast MRI from 1/1/12 to 12/31/15, 73 patients had 18F-FDG PET/CT evaluation performed within 1 week of breast MRI and no suspicious findings in the contralateral breast. MRI was performed on a 1.5T or 3.0T system. The imaging sequence included a triplane localizing sequence followed by sagittal fat-suppressed T2 -weighted sequence, and a bilateral sagittal T1 -weighted fat-suppressed fast spoiled gradient-echo sequence, which was performed before and three times after a rapid bolus injection (gadobenate dimeglumine, Multihance; Bracco Imaging; 0.1 mmol/kg) delivered through an IV catheter. The unaffected contralateral breast in these 73 patients underwent BPE and BPU assessments. For PET/CT BPU calculation, a 3D region of interest (ROI) was drawn around the glandular breast tissue and the maximum standardized uptake value (SUVmax ) was determined. Qualitative MRI BPE assessments were performed on a 4-point scale, in accordance with BI-RADS categories. Additional 3D quantitative MRI BPE analysis was performed using a previously published in-house technique. Spearman's correlation test and linear regression analysis was performed (SPSS, v. 24). RESULT: The median time interval between breast MRI and 18F-FDG PET/CT evaluation was 3 days (range, 0-6 days). BPU SUVmax mean value was 1.6 (SD, 0.53). Minimum and maximum BPU SUVmax values were 0.71 and 4.0. The BPU SUVmax values significantly correlated with both the qualitative and quantitative measurements of BPE, respectively (r(71) = 0.59, P < 0.001 and r(71) = 0.54, P < 0.001). Qualitatively assessed high BPE group (BI-RADS 3/4) had significantly higher BPU SUVmax of 1.9 (SD = 0.44) compared to low BPE group (BI-RADS 1/2) with an average BPU SUVmax of 1.17 (SD = 0.32) (P < 0.001). On linear regression analysis, BPU SUVmax significantly predicted qualitative and quantitative measurements of BPE (ß = 1.29, t(71) = 3.88, P < 0.001 and ß = 19.52, t(71) = 3.88, P < 0.001). CONCLUSION: There is a significant association between breast BPU and BPE, measured both qualitatively and quantitatively. Increased breast cancer risk in patients with high MRI BPE could be due to elevated basal metabolic activity of the normal breast tissue, which may provide a susceptible environment for tumor growth. LEVEL OF EVIDENCE: 3 Technical Efficacy: Stage 3 J. Magn. Reson. Imaging 2018;47:753-759.

Beyond Wires and Seeds: Reflector-guided Breast Lesion Localization and Excision.

Purpose To evaluate outcomes of Savi Scout (Cianna Medical, Aliso Viejo, Calif) reflector-guided localization and excision of breast lesions by analyzing reflector placement, localization, and removal, along with target excision and rates of repeat excision (referred to as re-excision). Materials and Methods A single-institution retrospective review of 100 women who underwent breast lesion localization and excision by using the Savi Scout surgical guidance system from June 2015 to May 2016 was performed. By using image guidance 0-8 days before surgery, 123 nonradioactive, infrared-activated, electromagnetic wave reflectors were percutaneously inserted adjacent to or within 111 breast targets. Twenty patients had two or three reflectors placed for bracketing or for localizing multiple lesions, and when ipsilateral, they were placed as close as 2.6 cm apart. Target and reflector were localized intraoperatively by one of two breast surgeons who used a handpiece that emitted infrared light and electromagnetic waves. Radiographs of the specimen and pathologic analysis helped verify target and reflector removal. Target to reflector distance was measured on the mammogram and radiograph of the specimen, and reflector depth was measured on the mammogram. Pathologic analysis was reviewed. Re-excision rates and complications were recorded. By using statistics software, descriptive statistics were generated with 95% confidence intervals (CIs) calculated. Results By using sonographic (40 of 123; 32.5%; 95% CI: 24.9%, 41.2%) or mammographic (83 of 123; 67.5%; 95% CI: 58.8% 75.1%) guidance, 123 (100%; 95% CI: 96.4%, 100%) reflectors were placed. Mean mammographic target to reflector distance was 0.3 cm. All 123 (100%; 95% CI: 96.4%, 100%) targets and reflectors were excised. Pathologic analysis yielded 54 of 110 malignancies (49.1%; 95% CI: 39.9%, 58.3%; average, 1.0 cm; range, 0.1-5 cm), 32 high-risk lesions (29.1%; 95% CI: 21.4%, 38.2%), and 24 benign lesions (21.8%; 95% CI: 115.1%, 30.4%). Four of 54 malignant cases (7.4%; 95% CI: 2.4%, 18.1%) demonstrated margins positive for cancer that required re-excision. Five of 110 radiographs of the specimen (4.5%; 95% CI: 1.7%, 10.4%) demonstrated increased distance between the target and reflector distance of greater than 1.0 cm (range, 1.1-2.6 cm) compared with postprocedure mammogram the day of placement, three of five were associated with hematomas, two of five migrated without identifiable cause. No related postoperative complications were identified. Conclusion Savi Scout is an accurate, reliable method to localize and excise breast lesions with acceptable margin positivity and re-excision rates. Bracketing is possible with reflectors as close as 2.6 cm. Savi Scout overcomes many limitations of other localization methods, which warrants further study. © RSNA, 2017.

The Contact Zone: A Common Site of Tumor Recurrence in a Patient Who Underwent Skin-Sparing Mastectomy and Myocutaneous Flap Reconstruction.

Routine magnetic resonance imaging (MRI) screening is not typically warranted in asymptomatic patients with a history of breast cancer and myocutaneous flap reconstruction due to the rare incidence of local tumor recurrence. We present a case of recurrent invasive ductal carcinoma along the contact zone between the transverse rectus abdominis myocutaneous (TRAM) flap and the native breast tissue that was incidentally detected on a routine high-risk screening-MRI of the breast in an asymptomatic patient with a history of breast cancer.

Assessing improvement in detection of breast cancer with three-dimensional automated breast US in women with dense breast tissue: the SomoInsight Study.

PURPOSE: To determine improvement in breast cancer detection by using supplemental three-dimensional (3D) automated breast (AB) ultrasonography (US) with screening mammography versus screening mammography alone in asymptomatic women with dense breasts. MATERIALS AND METHODS: Institutional review board approval and written informed consent were obtained for this HIPAA-compliant study. The SomoInsight Study was an observational, multicenter study conducted between 2009 and 2011. A total of 15 318 women (mean age, 53.3 years ± 10 [standard deviation]; range, 25-94 years) presenting for screening mammography alone with heterogeneously (50%-75%) or extremely (>75%) dense breasts were included, regardless of further risk characterization, and were followed up for 1 year. Participants underwent screening mammography alone followed by an AB US examination; results were interpreted sequentially. McNemar test was used to assess differences in cancer detection. RESULTS: Breast cancer was diagnosed at screening in 112 women: 82 with screening mammography and an additional 30 with AB US. Addition of AB US to screening mammography yielded an additional 1.9 detected cancers per 1000 women screened (95% confidence interval [CI]: 1.2, 2.7; P < .001). Of cancers detected with screening mammography, 62.2% (51 of 82) were invasive versus 93.3% (28 of 30) of additional cancers detected with AB US (P = .001). Of the 82 cancers detected with either screening mammography alone or the combined read, 17 were detected with screening mammography alone. Of these, 64.7% (11 of 17) were ductal carcinoma in situ versus 6.7% (two of 30) of cancers detected with AB US alone. Sensitivity for the combined read increased by 26.7% (95% CI: 18.3%, 35.1%); the increase in the recall rate per 1000 women screened was 284.9 (95% CI: 278.0, 292.2; P < .001). CONCLUSION: Addition of AB US to screening mammography in a generalizable cohort of women with dense breasts increased the cancer detection yield of clinically important cancers, but it also increased the number of false-positive results.

Evaluation of a hydrogel based breast biopsy marker (HydroMARK®) as an alternative to wire and radioactive seed localization for non-palpable breast lesions.

BACKGROUND AND OBJECTIVES: HydroMARK® is a newly available biopsy marker for image-guided needle biopsies of non-palpable breast lesions. Objective was to determine if the marker could be utilized independently for lesion localization using intra-operative ultrasound alone. METHODS: A single institution retrospective review identified patients who underwent surgical excision of breast lesions after placement of the HydroMARK®. Endpoints included intra-operative visualization of the marker, successful excision of the lesion, and presence of the marker on specimen radiograph. RESULTS: The study included 31 lesions in 25 patients. Twenty-nine (93.6%) HydroMARKSs® were adequately visualized by intra-operative ultrasound. Intra-operative ultrasound without pre-operative placement of a localizing device was successful for localization in six cases (19.4%). Intra-operative difficulties were encountered in 16 of 31 (51.6%) procedures. This included either extrusion of the marker when the biopsy tract was transected in 14 (45.2%) cases or migration of the marker prior to the procedure in two (6.4%) cases. The marker was visualized on specimen radiograph in 15 (48.4%) cases. CONCLUSIONS: While intraoperative sonographic visibility was excellent, a large number of excisions were associated with extrusion of the marker. Modifications are needed to improve acceptability of this marker for intra-operative localization independent of pre-operative wire or seed localization.

Underestimation of atypical ductal hyperplasia at MRI-guided 9-gauge vacuum-assisted breast biopsy.

OBJECTIVE: The purposes of this study were to determine the frequency of diagnosis of atypical ductal hyperplasia (ADH) at MRI-guided 9-gauge vacuum-assisted breast biopsy and to assess the rate of underestimation of ADH at subsequent surgical excision. MATERIALS AND METHODS: We conducted a retrospective review of medical records of 237 lesions consecutively detected with MRI and then subjected to MRI-guided 9-gauge vacuum-assisted breast biopsy during a 33-month period. Underestimated ADH was defined as a lesion yielding ADH at vacuum-assisted biopsy and cancer at surgery. Scientific tables were used to calculate 95% CI. RESULTS: Histologic analysis of MRI-guided vacuum-assisted breast biopsy specimens yielded ADH without cancer in 15 (6%) of 237 lesions. Among 15 patients in whom vacuum-assisted breast biopsy yielded ADH, the median age was 52 years (range, 46-68 years). The median number of specimens obtained was nine (range, 8-18 lesions). Median MRI lesion diameter was 1.3 cm (range, 0.7-7.0 cm). Among 15 MRI lesions, 10 (67%) were nonmasslike enhancement and five (33%) were masses. Surgical excision was performed on 13 lesions. Surgical histologic findings were malignancy in five (38%) of the cases, all ductal carcinoma in situ; high-risk lesion in six (46%) of the cases, including ADH without other high-risk lesions (n = 2), ADH and lobular carcinoma in situ (LCIS) (n = 1), ADH, LCIS, and papilloma (n =1), ADH and papilloma (n = 1), and LCIS (n = 1); and benign in two (15%) of the cases. These data indicated an ADH underestimation rate of 38% (95% CI, 14-68%). CONCLUSION: ADH without cancer was encountered in 6% of MRI-guided 9-gauge vacuum-assisted breast biopsies. ADH at MRI-guided vacuum-assisted breast biopsy is an indication for surgical excision because of the high (38%) frequency of underestimation of these lesions.