Deep Learning to Simulate Contrast-enhanced Breast MRI of Invasive Breast Cancer.

Background There is increasing interest in noncontrast breast MRI alternatives for tumor visualization to increase the accessibility of breast MRI. Purpose To evaluate the feasibility and accuracy of generating simulated contrast-enhanced T1-weighted breast MRI scans from precontrast MRI sequences in biopsy-proven invasive breast cancer with use of deep learning. Materials and Methods Women with invasive breast cancer and a contrast-enhanced breast MRI examination that was performed for initial evaluation of the extent of disease between January 2015 and December 2019 at a single academic institution were retrospectively identified. A three-dimensional, fully convolutional deep neural network simulated contrast-enhanced T1-weighted breast MRI scans from five precontrast sequences (T1-weighted non-fat-suppressed [FS], T1-weighted FS, T2-weighted FS, apparent diffusion coefficient, and diffusion-weighted imaging). For qualitative assessment, four breast radiologists (with 3-15 years of experience) blinded to whether the method of contrast was real or simulated assessed image quality (excellent, acceptable, good, poor, or unacceptable), presence of tumor enhancement, and maximum index mass size by using 22 pairs of real and simulated contrast-enhanced MRI scans. Quantitative comparison was performed using whole-breast similarity and error metrics and Dice coefficient analysis of enhancing tumor overlap. Results Ninety-six MRI examinations in 96 women (mean age, 52 years ± 12 [SD]) were evaluated. The readers assessed all simulated MRI scans as having the appearance of a real MRI scan with tumor enhancement. Index mass sizes on real and simulated MRI scans demonstrated good to excellent agreement (intraclass correlation coefficient, 0.73-0.86; P < .001) without significant differences (mean differences, -0.8 to 0.8 mm; P = .36-.80). Almost all simulated MRI scans (84 of 88 [95%]) were considered of diagnostic quality (ratings of excellent, acceptable, or good). Quantitative analysis demonstrated strong similarity (structural similarity index, 0.88 ± 0.05), low voxel-wise error (symmetric mean absolute percent error, 3.26%), and Dice coefficient of enhancing tumor overlap of 0.75 ± 0.25. Conclusion It is feasible to generate simulated contrast-enhanced breast MRI scans with use of deep learning. Simulated and real contrast-enhanced MRI scans demonstrated comparable tumor sizes, areas of tumor enhancement, and image quality without significant qualitative or quantitative differences. © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by Slanetz in this issue. An earlier incorrect version appeared online. This article was corrected on January 17, 2023.

Comparing Mammographic Density Assessed by Digital Breast Tomosynthesis or Digital Mammography: The Breast Cancer Surveillance Consortium.

Background Consistency in reporting Breast Imaging Reporting and Data System (BI-RADS) breast density on mammograms is important because breast density is used for breast cancer risk assessment and is reported directly to women and clinicians to inform decisions about supplemental screening. Purpose To assess the consistency of BI-RADS density reporting between digital breast tomosynthesis (DBT) and digital mammography (DM) and evaluate density as a breast cancer risk factor when assessed using DM versus DBT. Materials and Methods The Breast Cancer Surveillance Consortium is a prospective cohort study of women undergoing mammography with DM or DBT. This secondary analysis included women aged 40-79 years who underwent at least two screening mammography examinations less than 36 months apart. Percentage agreement and κ statistic were estimated for pairs of BI-RADS density assessments. Cox proportional hazards regression was used to calculate hazard ratios (HRs) of breast density as a risk factor for invasive breast cancer. Results A total of 403 326 pairs of mammograms from 342 149 women were evaluated. There were no significant differences in breast density assessment in pairs consisting of one DM and one DBT examination (57 516 of 74 729 [77%]; κ = 0.64), two DM examinations (238 678 of 301 743 [79%]; κ = 0.67), and two DBT examinations (20 763 of 26 854 [77%]; κ = 0.65). Results were similar when restricting the analyses to pairs read by the same radiologist. The breast cancer HRs for breast density were similar for DM and DBT (P = .45 for interaction). The HRs for density acquired using DM and DBT, respectively, were 0.55 (95% CI: 0.49, 0.63) and 0.37 (95% CI: 0.21, 0.66) for almost entirely fat, 1.47 (95% CI: 1.37, 1.58) and 1.36 (95% CI: 1.02, 1.82) for heterogeneously dense, and 1.72 (95% CI: 1.54, 1.93) and 2.05 (95% CI: 1.25, 3.36) for extremely dense breasts. Conclusion Radiologist reporting of Breast Imaging Reporting and Data System density obtained with digital breast tomosynthesis did not differ from that obtained with digital mammography. © RSNA, 2021 Online supplemental material is available for this article.

Utility and Outcomes of Preoperative Screening Breast MRI for Planned Bilateral Prophylactic Mastectomy in High-Risk Patients.

BACKGROUND. There is a paucity of data and consensus guidelines on the utility of preoperative MRI for planned bilateral prophylactic mastectomy. OBJECTIVE. The purpose of this study was to evaluate the utility of breast MRI performed in high-risk patients for the indication of planned bilateral prophylactic mastectomy, with attention given to the diagnostic performance for breast cancer detection. A secondary aim was to assess the potential impact of breast MRI findings on the decision to perform sentinel lymph node biopsy at the time of prophylactic mastectomy. METHODS. A retrospective database review identified MRI examinations performed at an academic medical center from August 2003 to January 2020 for the indication of planned bilateral prophylactic mastectomy. Patient demographics, imaging findings, operative details, and pathology were recorded. BI-RADS category 1 and 2 assessments were considered negative examinations, and BI-RADS category 3, 4, and 5 assessments were considered positive examinations. Descriptive statistics and performance metrics were calculated. RESULTS. The final cohort included 53 patients (mean age, 45 years). Most (35/53; 66.0%) studies were baseline examinations. Of the 53 patients, 31 (58.5%) had negative MRI examinations and 22 (41.5%) had positive MRI examinations. MRI detected two malignancies (one invasive lobular carcinoma and one high-grade ductal carcinoma in situ), both of which were assessed as BI-RADS category 4. The patient with invasive lobular cancer underwent sentinel lymph node biopsy at the time of mastectomy, which showed metastasis. Breast MRI had sensitivity of 100.0% and specificity of 60.8% for overall breast cancer detection and sensitivity of 100.0% and specificity of 59.6% for invasive cancer detection. CONCLUSION. Preoperative MRI for planned bilateral prophylactic mastectomy detected all cancers, indicating a potential role for MRI in impacting surgical decision making. CLINICAL IMPACT. Given the high NPV for cancer, our results suggest that lymph node biopsy may be safely avoided in patients with a negative MRI examination. This is clinically relevant because sentinel nodes cannot be identified after mastectomy.

Mean Apparent Diffusion Coefficient Is a Sufficient Conventional Diffusion-weighted MRI Metric to Improve Breast MRI Diagnostic Performance: Results from the ECOG-ACRIN Cancer Research Group A6702 Diffusion Imaging Trial.

Background The Eastern Cooperative Oncology Group and American College of Radiology Imaging Network Cancer Research Group A6702 multicenter trial helped confirm the potential of diffusion-weighted MRI for improving differential diagnosis of suspicious breast abnormalities and reducing unnecessary biopsies. A prespecified secondary objective was to explore the relative value of different approaches for quantitative assessment of lesions at diffusion-weighted MRI. Purpose To determine whether alternate calculations of apparent diffusion coefficient (ADC) can help further improve diagnostic performance versus mean ADC values alone for analysis of suspicious breast lesions at MRI. Materials and Methods This prospective trial (ClinicalTrials.gov identifier: NCT02022579) enrolled consecutive women (from March 2014 to April 2015) with a Breast Imaging Reporting and Data System category of 3, 4, or 5 at breast MRI. All study participants underwent standardized diffusion-weighted MRI (b = 0, 100, 600, and 800 sec/mm2). Centralized ADC measures were performed, including manually drawn whole-lesion and hotspot regions of interest, histogram metrics, normalized ADC, and variable b-value combinations. Diagnostic performance was estimated by using the area under the receiver operating characteristic curve (AUC). Reduction in biopsy rate (maintaining 100% sensitivity) was estimated according to thresholds for each ADC metric. Results Among 107 enrolled women, 81 lesions with outcomes (28 malignant and 53 benign) in 67 women (median age, 49 years; interquartile range, 41-60 years) were analyzed. Among ADC metrics tested, none improved diagnostic performance versus standard mean ADC (AUC, 0.59-0.79 vs AUC, 0.75; P = .02-.84), and maximum ADC had worse performance (AUC, 0.52; P < .001). The 25th-percentile ADC metric provided the best performance (AUC, 0.79; 95% CI: 0.70, 0.88), and a threshold using median ADC provided the greatest reduction in biopsy rate of 23.9% (95% CI: 14.8, 32.9; 16 of 67 BI-RADS category 4 and 5 lesions). Nonzero minimum b value (100, 600, and 800 sec/mm2) did not improve the AUC (0.74; P = .28), and several combinations of two b values (0 and 600, 100 and 600, 0 and 800, and 100 and 800 sec/mm2; AUC, 0.73-0.76) provided results similar to those seen with calculations of four b values (AUC, 0.75; P = .17-.87). Conclusion Mean apparent diffusion coefficient calculated with a two-b-value acquisition is a simple and sufficient diffusion-weighted MRI metric to augment diagnostic performance of breast MRI compared with more complex approaches to apparent diffusion coefficient measurement. © RSNA, 2020 Online supplemental material is available for this article.

Breast MRI during Neoadjuvant Chemotherapy: Lack of Background Parenchymal Enhancement Suppression and Inferior Treatment Response.

Background Suppression of background parenchymal enhancement (BPE) is commonly observed after neoadjuvant chemotherapy (NAC) at contrast-enhanced breast MRI. It was hypothesized that nonsuppressed BPE may be associated with inferior response to NAC. Purpose To investigate the relationship between lack of BPE suppression and pathologic response. Materials and Methods A retrospective review was performed for women with menopausal status data who were treated for breast cancer by one of 10 drug arms (standard NAC with or without experimental agents) between May 2010 and November 2016 in the Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and Molecular Analysis 2, or I-SPY 2 TRIAL (NCT01042379). Patients underwent MRI at four points: before treatment (T0), early treatment (T1), interregimen (T2), and before surgery (T3). BPE was quantitatively measured by using automated fibroglandular tissue segmentation. To test the hypothesis effectively, a subset of examinations with BPE with high-quality segmentation was selected. BPE change from T0 was defined as suppressed or nonsuppressed for each point. The Fisher exact test and the Z tests of proportions with Yates continuity correction were used to examine the relationship between BPE suppression and pathologic complete response (pCR) in hormone receptor (HR)-positive and HR-negative cohorts. Results A total of 3528 MRI scans from 882 patients (mean age, 48 years ± 10 [standard deviation]) were reviewed and the subset of patients with high-quality BPE segmentation was determined (T1, 433 patients; T2, 396 patients; T3, 380 patients). In the HR-positive cohort, an association between lack of BPE suppression and lower pCR rate was detected at T2 (nonsuppressed vs suppressed, 11.8% [six of 51] vs 28.9% [50 of 173]; difference, 17.1% [95% CI: 4.7, 29.5]; P = .02) and T3 (nonsuppressed vs suppressed, 5.3% [two of 38] vs 27.4% [48 of 175]; difference, 22.2% [95% CI: 10.9, 33.5]; P = .003). In the HR-negative cohort, patients with nonsuppressed BPE had lower estimated pCR rate at all points, but the P values for the association were all greater than .05. Conclusions In hormone receptor-positive breast cancer, lack of background parenchymal enhancement suppression may indicate inferior treatment response. © RSNA, 2021 Online supplemental material is available for this article. See also the editorial by Philpotts in this issue.

Denoising and Multiple Tissue Compartment Visualization of Multi-b-Valued Breast Diffusion MRI.

BACKGROUND: Multi-b-valued/multi-shell diffusion provides potentially valuable metrics in breast MRI but suffers from low signal-to-noise ratio and has potentially long scan times. PURPOSE: To investigate the effects of model-based denoising with no loss of spatial resolution on multi-shell breast diffusion MRI; to determine the effects of downsampling on multi-shell diffusion; and to quantify these effects in multi-b-valued (three directions per b-value) acquisitions. STUDY TYPE: Prospective ("fully-sampled" multi-shell) and retrospective longitudinal (multi-b). SUBJECTS: One normal subject (multi-shell) and 10 breast cancer subjects imaging at four timepoints (multi-b). FIELD STRENGTH/SEQUENCE: 3T multi-shell acquisition and 1.5T multi-b acquisition. ASSESSMENT: The "fully-sampled" multi-shell acquisition was retrospectively downsampled to determine the bias and error from downsampling. Mean, axial/parallel, radial diffusivity, and fractional anisotropy (FA) were analyzed. Denoising was applied retrospectively to the multi-b-valued breast cancer subject dataset and assessed subjectively for image noise level and tumor conspicuity. STATISTICAL TESTS: Parametric paired t-test (P < 0.05 considered statistically significant) on mean and coefficient of variation of each metric-the apparent diffusion coefficient (ADC) from all b-values, fast ADC, slow ADC, and perfusion fraction. Paired and two-sample t-tests for each metric comparing normal and tumor tissue. RESULTS: In the multi-shell data, denoising effectively suppressed FA (-45% to -78%), with small biases in mean diffusivity (-5% in normal, +23% in tumor, and -4% in vascular compartments). In the multi-b data, denoising resulted in small biases to the ADC metrics in tumor and normal contralateral tissue (by -3% to +11%), but greatly reduced the coefficient of variation for every metric (by -1% to -24%). Denoising improved differentiation of tumor and normal tissue regions in most metrics and timepoints; subjectively, image noise level and tumor conspicuity were improved in the fast ADC maps. DATA CONCLUSION: Model-based denoising effectively suppressed erroneously high FA and improved the accuracy of diffusivity metrics. EVIDENCE LEVEL: 3 TECHNICAL EFFICACY STAGE: 1.

Frequency and Outcomes of New Suspicious Lesions on Breast MRI in the Setting of Neoadjuvant Therapy.

OBJECTIVE. The purpose of this article was to determine the frequency and outcomes of new suspicious findings on breast MRI after initiation of neoadjuvant therapy. MATERIALS AND METHODS. A retrospective database review identified all breast MRI examinations performed to assess response to neoadjuvant therapy between 2010 and 2018. Cases with new suspicious lesions assessed as BI-RADS 4 or 5 and found after the initiation of neoadjuvant treatment were included. Cases with no pretreatment MRI, cases in which the suspicious lesion was present on the baseline MRI but remained suspicious, and cases with insufficient follow-up were excluded. Radiologic, pathologic, and surgical reports were reviewed. Malignant outcomes were determined by pathologic examination. Benignity was established by pathologic examination, follow-up imaging, or both. A total of 419 breast MRI examinations in 297 women were performed to assess response to neoadjuvant therapy. After exclusions, 23 MRI examinations (5.5%) with new suspicious findings, all assessed as BI-RADS 4, comprised the final cohort. RESULTS. Of the 23 lesions, 13 new suspicious findings (56.5%) were contralateral to the known malignancy, nine (39.1%) were ipsilateral, and one (4.3%) involved the bilateral breasts. Lesion types included mass (16, 69.6%), nonmass enhancement (5, 21.7%) and focus (2, 8.7%). None of the new suspicious findings were malignant. CONCLUSION. New suspicious findings occurred in 5.5% of breast MRI examinations performed to monitor response to neoadjuvant therapy, and none of these new lesions were malignant. Our findings suggest that new lesions that arise in the setting of neoadjuvant therapy are highly unlikely to represent a new site of malignancy, particularly if the index malignancy shows treatment response. Larger studies are needed to confirm whether biopsy may be safely averted in this scenario.

Mask-Guided Convolutional Neural Network for Breast Tumor Prognostic Outcome Prediction on 3D DCE-MR Images.

In this proof-of-concept work, we have developed a 3D-CNN architecture that is guided by the tumor mask for classifying several patient-outcomes in breast cancer from the respective 3D dynamic contrast-enhanced MRI (DCE-MRI) images. The tumor masks on DCE-MRI images were generated using pre- and post-contrast images and validated by experienced radiologists. We show that our proposed mask-guided classification has a higher accuracy than that from either the full image without tumor masks (including background) or the masked voxels only. We have used two patient outcomes for this study: (1) recurrence of cancer after 5 years of imaging and (2) HER2 status, for comparing accuracies of different models. By looking at the activation maps, we conclude that an image-based prediction model using 3D-CNN could be improved by even a conservatively generated mask, rather than overly trusting an unguided, blind 3D-CNN. A blind CNN may classify accurately enough, while its attention may really be focused on a remote region within 3D images. On the other hand, only using a conservatively segmented region may not be as good for classification as using full images but forcing the model's attention toward the known regions of interest.

US as the Primary Imaging Modality in the Evaluation of Palpable Breast Masses in Breastfeeding Women, Including Those of Advanced Maternal Age.

Background Women are increasingly delaying childbearing, and thus lactation, into their 30s and 40s, when mammography would typically be the initial imaging modality to evaluate palpable masses in the general population. Current guidelines recommend US as the first-line imaging modality for palpable masses in pregnant and lactating women, but data regarding breastfeeding women age 30 years and older are near nonexistent. Purpose To evaluate the diagnostic performance of targeted US as the primary imaging modality for the evaluation of palpable masses in lactating women, including those of advanced maternal age. Materials and Methods Lactating women with palpable breast masses evaluated at targeted US over a 17-year period (January 2000 to July 2017) were retrospectively identified. All US evaluations were performed at diagnostic evaluation, and mammography was performed at the discretion of the interpreting radiologist. Breast Imaging Reporting and Data System assessments, imaging, and pathology results were collected. Descriptive statistics and 2 × 2 contingency tables were assessed at the patient level. Results There were 167 women (mean age, 35 years ± 5 [standard deviation]), 101 of whom (60%) were of advanced maternal age (≥35 years). All women underwent targeted US, and 98 (59%) underwent mammography in addition to US. The frequency of malignancy was five of 167 (3.0%). Targeted US demonstrated a sensitivity and specificity of five of five (100%; 95% confidence interval [CI]: 48%, 100%) and 114 of 162 (70%; 95% CI: 63%, 77%), respectively. Negative predictive value, positive predictive value of an abnormal examination, and positive predictive value of biopsy were 114 of 114 (100%; 95% CI: 97%, 100%), five of 53 (9.4%; 95% CI: 3%, 21%), and five of 50 (10%; 95% CI: 3%, 22%), respectively. In the subset of 98 women who underwent mammography in addition to US, mammography depicted seven incidental suspicious findings, which lowered the specificity from 62 of 93 (67%; 95% CI: 56%, 76%) to 57 of 93 (61%; 95% CI: 51%, 71%) (P = .02). Conclusion Targeted US depicted all malignancies in lactating women with palpable masses. Adding mammography increased false-positive findings without any additional cancer diagnoses. © RSNA, 2020 See also the editorial by Newell in this issue.

Role of Breast MRI in the Evaluation and Detection of DCIS: Opportunities and Challenges.

Historically, breast magnetic resonance imaging (MRI) was not considered an effective modality in the evaluation of ductal carcinoma in situ (DCIS). Over the past decade this has changed, with studies demonstrating that MRI is the most sensitive imaging tool for detection of all grades of DCIS. It has been suggested that not only is breast MRI the most sensitive imaging tool for detection but it may also detect the most clinically relevant DCIS lesions. The role and outcomes of MRI in the preoperative setting for patients with DCIS remains controversial; however, several studies have shown benefit in the preoperative evaluation of extent of disease as well as predicting an underlying invasive component. The most common presentation of DCIS on MRI is nonmass enhancement (NME) in a linear or segmental distribution pattern. Maximizing breast MRI spatial resolution is therefore beneficial, given the frequent presentation of DCIS as NME on MRI. Emerging MRI techniques, such as diffusion-weighted imaging (DWI), have shown promising potential to discriminate DCIS from benign and invasive lesions. Future opportunities including advanced imaging visual techniques, radiomics/radiogenomics, and machine learning / artificial intelligence may also be applicable to the detection and treatment of DCIS. Level of Evidence: 3 Technical Efficacy Stage: 3 J. Magn. Reson. Imaging 2019. J. Magn. Reson. Imaging 2020;52:697-709.

The Role of Breast MRI in Detecting Asymptomatic Recurrence After Therapeutic Mastectomy.

OBJECTIVE. MRI is not routinely used to screen for cancer recurrence after therapeutic mastectomy; however, data on this topic are sparse. We performed this study to determine the utility of breast MRI in detecting asymptomatic locoregional recurrence after therapeutic mastectomy. MATERIALS AND METHODS. A retrospective record review identified all breast MRI studies performed in women who had undergone unilateral therapeutic mastectomy over a 6-year period (January 1, 2010, to January 1, 2016). A total of 402 studies were performed in 191 women between the ages of 26 and 78 years old, none of whom were experiencing symptoms on the mastectomy side. BI-RADS assessments for the mastectomy side were extracted from the radiology reports, and the electronic medical records were reviewed for surgical and oncologic history, clinical and imaging follow-up, and pathologic results. Malignancy was determined by pathologic results. Benignity was confirmed by at least one of the following: pathologic results, at least 12 months of documented disease-free clinical follow-up, or at least 12 months of documented disease-free imaging follow-up. Descriptive statistical and 2 × 2 contingency table analyses were performed. RESULTS. In all, 395 MR images (98.3%) were assessed as showing benign findings on the mastectomy side. Seven (1.7%) were interpreted as showing positive findings on the mastectomy side (BI-RADS category 4, suspicious for malignancy). Biopsy was performed in four of the seven positive interpretations. All four biopsies yielded malignancy for a positive predictive value of biopsy of 100%. The three remaining positive cases did not include biopsy; however, in each case, follow-up imaging showed improvement or resolution of the finding, yielding a positive predictive value of an abnormal examination of 57.1%. Two MRI studies were false-negative, with local recurrence within 12 months after MRI deemed to show benign findings, yielding a negative predictive value of 99.5%. Sensitivity and specificity were 66.7% and 99.2%, respectively. The cancer detection rate in the asymptomatic mastectomy side for all MRI examinations was 10 cancers per 1000 examinations. CONCLUSION. Our findings support inclusion of the mastectomy side in MRI examinations of the contralateral breast to screen for cancer recurrence after therapeutic mastectomy.

Utility of Diagnostic Mammography as the Primary Imaging Modality for Palpable Lumps in Women With Almost Entirely Fatty Breasts.

OBJECTIVE. The purpose of this study was to assess the performance of diagnostic mammography alone for evaluation of palpable symptoms in women with almost entirely fatty breast composition. MATERIALS AND METHODS. All diagnostic mammograms performed for palpable symptoms in women who had been assigned a breast density of "almost entirely fatty" over an 8-year period (2009-2017) at an academic breast center were retrospectively identified. Each symptomatic breast was considered a separate case and analyses were performed at the case level. Clinical, imaging, and pathologic results were reviewed. Descriptive statistics and 2 × 2 contingency table analyses were performed. RESULTS. The study cohort included 323 cases evaluated with mammography. Of these, 294 (91%) had undergone targeted ultrasound. At mammography, 240 (74%) had no correlate to the palpable lump; 38 (12%), a benign correlate; and 45 (14%), a suspicious correlate. Three cases had incidental suspicious mammographic findings, for a total of 48 positive mammography cases. Twenty-seven (8%) cases were malignant. Mammography alone detected all but one cancer, which was detected by ultrasound. In retrospect, the woman from whom this single false-negative mammogram was obtained did not have almost entirely fatty breast density. Mammography alone yielded a negative predictive value of 99.6%, percentage of diagnostic examinations recommended for biopsy that resulted in a tissue diagnosis of malignancy within 1 year of 54%, sensitivity of 96%, and specificity of 93%. Adjunct ultrasound contributed to 11 false-positives but also identified benign correlates in eight cases with no mammographic finding. CONCLUSION. In patients with almost entirely fatty breast tissue presenting with palpable symptoms, mammography alone had a high sensitivity and specificity. Our results support that mammography alone may be sufficient for evaluation of palpable symptoms in these women as long as density criteria are strictly applied.

An Open-Source, Vender Agnostic Hardware and Software Pipeline for Integration of Artificial Intelligence in Radiology Workflow.

Although machine learning (ML) has made significant improvements in radiology, few algorithms have been integrated into clinical radiology workflow. Complex radiology IT environments and Picture Archiving and Communication System (PACS) pose unique challenges in creating a practical ML schema. However, clinical integration and testing are critical to ensuring the safety and accuracy of ML algorithms. This study aims to propose, develop, and demonstrate a simple, efficient, and understandable hardware and software system for integrating ML models into the standard radiology workflow and PACS that can serve as a framework for testing ML algorithms. A Digital Imaging and Communications in Medicine/Graphics Processing Unit (DICOM/GPU) server and software pipeline was established at a metropolitan county hospital intranet to demonstrate clinical integration of ML algorithms in radiology. A clinical ML integration schema, agnostic to the hospital IT system and specific ML models/frameworks, was implemented and tested with a breast density classification algorithm and prospectively evaluated for time delays using 100 digital 2D mammograms. An open-source clinical ML integration schema was successfully implemented and demonstrated. This schema allows for simple uploading of custom ML models. With the proposed setup, the ML pipeline took an average of 26.52 s per second to process a batch of 100 studies. The most significant processing time delays were noted in model load and study stability times. The code is made available at " http://bit.ly/2Z121hX ". We demonstrated the feasibility to deploy and utilize ML models in radiology without disrupting existing radiology workflow.

Combined Benefit of Quantitative Three-Compartment Breast Image Analysis and Mammography Radiomics in the Classification of Breast Masses in a Clinical Data Set.

Purpose To investigate the combination of mammography radiomics and quantitative three-compartment breast (3CB) image analysis of dual-energy mammography to limit unnecessary benign breast biopsies. Materials and Methods For this prospective study, dual-energy craniocaudal and mediolateral oblique mammograms were obtained immediately before biopsy in 109 women (mean age, 51 years; range, 31-85 years) with Breast Imaging Reporting and Data System category 4 or 5 breast masses (35 invasive cancers, 74 benign) from 2013 through 2017. The three quantitative compartments of water, lipid, and protein thickness at each pixel were calculated from the attenuation at high and low energy by using a within-image phantom. Masses were automatically segmented and features were extracted from the low-energy mammograms and the quantitative compartment images. Tenfold cross-validations using a linear discriminant classifier with predefined feature signatures helped differentiate between malignant and benign masses by means of (a) water-lipid-protein composition images alone, (b) mammography radiomics alone, and (c) a combined image analysis of both. Positive predictive value of biopsy performed (PPV3) at maximum sensitivity was the primary performance metric, and results were compared with those for conventional diagnostic digital mammography. Results The PPV3 for conventional diagnostic digital mammography in our data set was 32.1% (35 of 109; 95% confidence interval [CI]: 23.9%, 41.3%), with a sensitivity of 100%. In comparison, combined mammography radiomics plus quantitative 3CB image analysis had PPV3 of 49% (34 of 70; 95% CI: 36.5%, 58.9%; P < .001), with a sensitivity of 97% (34 of 35; 95% CI: 90.3%, 100%; P < .001) and 35.8% (39 of 109) fewer total biopsies (P < .001). Conclusion Quantitative three-compartment breast image analysis of breast masses combined with mammography radiomics has the potential to reduce unnecessary breast biopsies. © RSNA, 2018 Online supplemental material is available for this article.

Test-retest repeatability and reproducibility of ADC measures by breast DWI: Results from the ACRIN 6698 trial.

BACKGROUND: Quantitative diffusion-weighted imaging (DWI) MRI is a promising technique for cancer characterization and treatment monitoring. Knowledge of the reproducibility of DWI metrics in breast tumors is necessary to apply DWI as a clinical biomarker. PURPOSE: To evaluate the repeatability and reproducibility of breast tumor apparent diffusion coefficient (ADC) in a multi-institution clinical trial setting, using standardized DWI protocols and quality assurance (QA) procedures. STUDY TYPE: Prospective. SUBJECTS: In all, 89 women from nine institutions undergoing neoadjuvant chemotherapy for invasive breast cancer. FIELD STRENGTH/SEQUENCE: DWI was acquired before and after patient repositioning using a four b-value, single-shot echo-planar sequence at 1.5T or 3.0T. ASSESSMENT: A QA procedure by trained operators assessed artifacts, fat suppression, and signal-to-noise ratio, and determine study analyzability. Mean tumor ADC was measured via manual segmentation of the multislice tumor region referencing DWI and contrast-enhanced images. Twenty cases were evaluated multiple times to assess intra- and interoperator variability. Segmentation similarity was assessed via the Sørenson-Dice similarity coefficient. STATISTICAL TESTS: Repeatability and reproducibility were evaluated using within-subject coefficient of variation (wCV), intraclass correlation coefficient (ICC), agreement index (AI), and repeatability coefficient (RC). Correlations were measured by Pearson's correlation coefficients. RESULTS: In all, 71 cases (80%) passed QA evaluation: 44 at 1.5T, 27 at 3.0T; 60 pretreatment, 11 after 3 weeks of taxane-based treatment. ADC repeatability was excellent: wCV = 4.8% (95% confidence interval [CI] 4.0, 5.7%), ICC = 0.97 (95% CI 0.95, 0.98), AI = 0.83 (95% CI 0.76, 0.87), and RC = 0.16 * 10-3 mm2 /sec (95% CI 0.13, 0.19). The results were similar across field strengths and timepoint subgroups. Reproducibility was excellent: interreader ICC = 0.92 (95% CI 0.80, 0.97) and intrareader ICC = 0.91 (95% CI 0.78, 0.96). DATA CONCLUSION: Breast tumor ADC can be measured with excellent repeatability and reproducibility in a multi-institution setting using a standardized protocol and QA procedure. Improvements to DWI image quality could reduce loss of data in clinical trials. LEVEL OF EVIDENCE: 2 Technical Efficacy: Stage 1 J. Magn. Reson. Imaging 2019;49:1617-1628.

Utility and Outcomes of Imaging Evaluation for Palpable Lumps in the Postmastectomy Patient.

OBJECTIVE. The objective of our study was to assess the utility of targeted breast ultrasound and mammography in evaluating palpable lumps in the mastectomy bed. MATERIALS AND METHODS. This retrospective study identified postmastectomy patients who presented for initial imaging evaluation of palpable lumps between January 2009 and December 2015. Clinical, imaging, and pathology results were reviewed. Surgical reconstruction type and percutaneous sampling data were collected. Patients were excluded if they had known malignancy at imaging presentation, if the palpable lump was not at the mastectomy site, or if there was less than 1 year clinical or imaging follow-up in the absence of biopsy. Each palpable site was assigned as a case, and analyses were performed at the case level. RESULTS. Among the 101 patients with a history of prophylactic or therapeutic mastectomy who presented during the study period, 118 palpable cases met the inclusion criteria. All 118 cases were evaluated with ultrasound and 43 with mammography. Among the 75 cases evaluated with ultrasound alone, nine cancers were detected. Among the 43 cases evaluated with both ultrasound and mammography, three cancers were sonographically detected, of which two were mammographically visible and one was mammographically occult. There were two false-negative ultrasound cases; both underwent sampling because of the level of clinical suspicion. In total, 14 palpable lumps in 12 patients were malignant, and 104 palpable lumps in 89 patients were nonmalignant. Targeted ultrasound yielded a negative predictive value (NPV) of 97% and a positive predictive value 2 of 27%. CONCLUSION. Our data suggest that targeted breast ultrasound, with its high NPV, should be the initial imaging test of choice for palpable lumps after mastectomy. Mammography yielded no additional cancers but was helpful in confirming benign diagnoses. The two false-negative ultrasound cases support palpation-guided sampling for imaging-occult and clinically suspicious palpable lumps.

Large Scale Semi-Automated Labeling of Routine Free-Text Clinical Records for Deep Learning.

Breast cancer is a leading cause of cancer death among women in the USA. Screening mammography is effective in reducing mortality, but has a high rate of unnecessary recalls and biopsies. While deep learning can be applied to mammography, large-scale labeled datasets, which are difficult to obtain, are required. We aim to remove many barriers of dataset development by automatically harvesting data from existing clinical records using a hybrid framework combining traditional NLP and IBM Watson. An expert reviewer manually annotated 3521 breast pathology reports with one of four outcomes: left positive, right positive, bilateral positive, negative. Traditional NLP techniques using seven different machine learning classifiers were compared to IBM Watson's automated natural language classifier. Techniques were evaluated using precision, recall, and F-measure. Logistic regression outperformed all other traditional machine learning classifiers and was used for subsequent comparisons. Both traditional NLP and Watson's NLC performed well for cases under 1024 characters with weighted average F-measures above 0.96 across all classes. Performance of traditional NLP was lower for cases over 1024 characters with an F-measure of 0.83. We demonstrate a hybrid framework using traditional NLP techniques combined with IBM Watson to annotate over 10,000 breast pathology reports for development of a large-scale database to be used for deep learning in mammography. Our work shows that traditional NLP and IBM Watson perform extremely well for cases under 1024 characters and can accelerate the rate of data annotation.

Qualitative Radiogenomics: Association between Oncotype DX Test Recurrence Score and BI-RADS Mammographic and Breast MR Imaging Features.

Purpose To evaluate the association between Breast Imaging Reporting and Data System (BI-RADS) mammographic and magnetic resonance (MR) imaging features and breast cancer recurrence risk in patients with estrogen receptor-positive breast cancer who underwent the Oncotype DX assay. Materials and Methods In this institutional review board-approved and HIPAA-compliant protocol, 408 patients diagnosed with invasive breast cancer between 2004 and 2013 who underwent the Oncotype DX assay were identified. Mammographic and MR imaging features were retrospectively collected according to the BI-RADS lexicon. Linear regression assessed the association between imaging features and Oncotype DX test recurrence score (ODxRS), and post hoc pairwise comparisons assessed ODxRS means by using imaging features. Results Mammographic breast density was inversely associated with ODxRS (P ≤ .05). Average ODxRS for density category A was 24.4 and that for density category D was 16.5 (P < .02). Both indistinct mass margins and fine linear branching calcifications at mammography were significantly associated with higher ODxRS (P < .01 and P < .03, respectively). Masses with indistinct margins had an average ODxRS of 31.3, which significantly differed from the ODxRS of 18.5 for all other mass margins (P < .01). The average ODxRS for fine linear branching calcifications was 29.6, whereas the ODxRS for all other suspicious calcification morphologies was 19.4 (P < .03). Average ODxRS was significantly higher for irregular mass margins at MR imaging compared with spiculated mass margins (24.0 vs 17.6; P < .02). The presence of nonmass enhancement at MR imaging was associated with lower ODxRS than was its absence (16.4 vs 19.9; P < .05). Conclusion The BI-RADS features of mammographic breast density, calcification morphology, mass margins at mammography and MR imaging, and nonmass enhancement at MR imaging have the potential to serve as imaging biomarkers of breast cancer recurrence risk. Further prospective studies involving larger patient cohorts are needed to validate these preliminary findings. © RSNA, 2017 Online supplemental material is available for this article.