Breast MRI Background Parenchymal Enhancement Categorization Using Deep Learning: Outperforming the Radiologist.

BACKGROUND: Background parenchymal enhancement (BPE) is assessed on breast MRI reports as mandated by the Breast Imaging Reporting and Data System (BI-RADS) but is prone to inter and intrareader variation. Semiautomated and fully automated BPE assessment tools have been developed but none has surpassed radiologist BPE designations. PURPOSE: To develop a deep learning model for automated BPE classification and to compare its performance with current standard-of-care radiology report BPE designations. STUDY TYPE: Retrospective. POPULATION: Consecutive high-risk patients (i.e. >20% lifetime risk of breast cancer) who underwent contrast-enhanced screening breast MRI from October 2013 to January 2019. The study included 5224 breast MRIs, divided into 3998 training, 444 validation, and 782 testing exams. On radiology reports, 1286 exams were categorized as high BPE (i.e., marked or moderate) and 3938 as low BPE (i.e., mild or minimal). FIELD STRENGTH/SEQUENCE: A 1.5 T or 3 T system; one precontrast and three postcontrast phases of fat-saturated T1-weighted dynamic contrast-enhanced imaging. ASSESSMENT: Breast MRIs were used to develop two deep learning models (Slab artificial intelligence (AI); maximum intensity projection [MIP] AI) for BPE categorization using radiology report BPE labels. Models were tested on a heldout test sets using radiology report BPE and three-reader averaged consensus as the reference standards. STATISTICAL TESTS: Model performance was assessed using receiver operating characteristic curve analysis. Associations between high BPE and BI-RADS assessments were evaluated using McNemar's chi-square test (α* = 0.025). RESULTS: The Slab AI model significantly outperformed the MIP AI model across the full test set (area under the curve of 0.84 vs. 0.79) using the radiology report reference standard. Using three-reader consensus BPE labels reference standard, our AI model significantly outperformed radiology report BPE labels. Finally, the AI model was significantly more likely than the radiologist to assign "high BPE" to suspicious breast MRIs and significantly less likely than the radiologist to assign "high BPE" to negative breast MRIs. DATA CONCLUSION: Fully automated BPE assessments for breast MRIs could be more accurate than BPE assessments from radiology reports. LEVEL OF EVIDENCE: 4 TECHNICAL EFFICACY STAGE: 3.

Diagnostic value of radiomics and machine learning with dynamic contrast-enhanced magnetic resonance imaging for patients with atypical ductal hyperplasia in predicting malignant upgrade.

PURPOSE: To investigate whether radiomics features extracted from magnetic resonance imaging (MRI) of patients with biopsy-proven atypical ductal hyperplasia (ADH) coupled with machine learning can differentiate high-risk lesions that will upgrade to malignancy at surgery from those that will not, and to determine if qualitatively and semi-quantitatively assessed imaging features, clinical factors, and image-guided biopsy technical factors are associated with upgrade rate. METHODS: This retrospective study included 127 patients with 139 breast lesions yielding ADH at biopsy who were assessed with multiparametric MRI prior to biopsy. Two radiologists assessed all lesions independently and with a third reader in consensus according to the BI-RADS lexicon. Univariate analysis and multivariate modeling were performed to identify significant radiomic features to be included in a machine learning model to discriminate between lesions that upgraded to malignancy on surgery from those that did not. RESULTS: Of 139 lesions, 28 were upgraded to malignancy at surgery, while 111 were not upgraded. Diagnostic accuracy was 53.6%, specificity 79.2%, and sensitivity 15.3% for the model developed from pre-contrast features, and 60.7%, 86%, and 22.8% for the model developed from delta radiomics datasets. No significant associations were found between any radiologist-assessed lesion parameters and upgrade status. There was a significant correlation between the number of specimens sampled during biopsy and upgrade status (p = 0.003). CONCLUSION: Radiomics analysis coupled with machine learning did not predict upgrade status of ADH. The only significant result from this analysis is between the number of specimens sampled during biopsy procedure and upgrade status at surgery.

Can Follow-up be Avoided for Probably Benign US Masses with No Enhancement on MRI?

OBJECTIVES: To assess whether no enhancement on pre-treatment MRI can rule out malignancy of additional US mass(es) initially assessed as BI-RADS 3 or 4 in women with newly diagnosed breast cancer. METHODS: This retrospective study included consecutive women from 2010-2018 with newly diagnosed breast cancer; at least one additional breast mass (distinct from index cancer) assigned a BI-RADS 3 or 4 on US; and a bilateral contrast-enhanced breast MRI performed within 90 days of US. All malignant masses were pathologically proven; benign masses were pathologically proven or defined as showing at least 2 years of imaging stability. Incidence of malignant masses and NPV were calculated on a per-patient level using proportions and exact 95% CIs. RESULTS: In 230 patients with 309 additional masses, 140/309 (45%) masses did not enhance while 169/309 (55%) enhanced on MRI. Of the 140 masses seen in 105 women (mean age, 54 years; range 28-82) with no enhancement on MRI, all had adequate follow-up and 140/140 (100%) were benign, of which 89/140 (63.6%) were pathologically proven and 51/140 (36.4%) demonstrated at least 2 years of imaging stability. Pre-treatment MRI demonstrating no enhancement of US mass correlate(s) had an NPV of 100% (95% CI 96.7-100.0). CONCLUSIONS: All BI-RADS 3 and 4 US masses with a non-enhancing correlate on pre-treatment MRI were benign. The incorporation of MRI, when ordered by the referring physician, may decrease unnecessary follow-up imaging and/or biopsy if the initial US BI-RADS assessment and management recommendation were to be retrospectively updated. KEY POINTS: • Of 309 BI-RADS 3 or 4 US masses with a corresponding mass on MRI, 140/309 (45%) demonstrated no enhancement whereas 169/309 (55%) demonstrated enhancement • All masses classified as BI-RADS 3 or 4 on US without enhancement on MRI were benign • MRI can rule out malignancy in non-enhancing US masses with an NPV of 100.

Diagnostic value of diffusion-weighted imaging with synthetic b-values in breast tumors: comparison with dynamic contrast-enhanced and multiparametric MRI.

OBJECTIVES: To assess DWI for tumor visibility and breast cancer detection by the addition of different synthetic b-values. METHODS: Eighty-four consecutive women who underwent a breast-multiparametric-MRI (mpMRI) with enhancing lesions on DCE-MRI (BI-RADS 2-5) were included in this IRB-approved retrospective study from September 2018 to March 2019. Three readers evaluated DW acquired b-800 and synthetic b-1000, b-1200, b-1500, and b-1800 s/mm2 images for lesion visibility and preferred b-value based on lesion conspicuity. Image quality (1-3 scores) and breast composition (BI-RADS) were also recorded. Diagnostic parameters for DWI were determined using a 1-5 malignancy score based on qualitative imaging parameters (acquired + preferred synthetic b-values) and ADC values. BI-RADS classification was used for DCE-MRI and quantitative ADC values + BI-RADS were used for mpMRI. RESULTS: Sixty-four malignant (average = 23 mm) and 39 benign (average = 8 mm) lesions were found in 80 women. Although b-800 achieved the best image quality score, synthetic b-values 1200-1500 s/mm2 were preferred for lesion conspicuity, especially in dense breast. b-800 and synthetic b-1000/b-1200 s/mm2 values allowed the visualization of 84-90% of cancers visible with DCE-MRI performing better than b-1500/b-1800 s/mm2. DWI was more specific (86.3% vs 65.7%, p < 0.001) but less sensitive (62.8% vs 90%, p < 0.001) and accurate (71% vs 80.7%, p = 0.003) than DCE-MRI for breast cancer detection, where mpMRI was the most accurate modality accounting for less false positive cases. CONCLUSION: The addition of synthetic b-values enhances tumor conspicuity and could potentially improve tumor visualization particularly in dense breast. However, its supportive role for DWI breast cancer detection is still not definite. KEY POINTS: • The addition of synthetic b-values (1200-1500 s/mm2) to acquired DWI afforded a better lesion conspicuity without increasing acquisition time and was particularly useful in dense breasts. • Despite the use of synthetic b-values, DWI was less sensitive and accurate than DCE-MRI for breast cancer detection. • A multiparametric MRI modality still remains the best approach having the highest accuracy for breast cancer detection and thus reducing the number of unnecessary biopsies.

MRI background parenchymal enhancement, fibroglandular tissue, and mammographic breast density in patients with invasive lobular breast cancer on adjuvant endocrine hormonal treatment: associations with survival.

BACKGROUND: To investigate if baseline and/or changes in contralateral background parenchymal enhancement (BPE) and fibroglandular tissue (FGT) measured on magnetic resonance imaging (MRI) and mammographic breast density (MD) can be used as imaging biomarkers for overall and recurrence-free survival in patients with invasive lobular carcinomas (ILCs) undergoing adjuvant endocrine treatment. METHODS: Women who fulfilled the following inclusion criteria were included in this retrospective HIPAA-compliant IRB-approved study: unilateral ILC, pre-treatment breast MRI and/or mammography from 2000 to 2010, adjuvant endocrine treatment, follow-up MRI, and/or mammography 1-2 years after treatment onset. BPE, FGT, and mammographic MD of the contralateral breast were independently graded by four dedicated breast radiologists according to BI-RADS. Associations between the baseline levels and change in levels of BPE, FGT, and MD with overall survival and recurrence-free survival were assessed using Kaplan-Meier survival curves and Cox regression analysis. RESULTS: Two hundred ninety-eight patients (average age = 54.1 years, range = 31-79) fulfilled the inclusion criteria. The average follow-up duration was 11.8 years (range = 2-19). Baseline and change in levels of BPE, FGT, and MD were not significantly associated with recurrence-free or overall survival. Recurrence-free and overall survival were affected by histological subtype (p < 0.0001), number of metastatic axillary lymph nodes (p < 0.0001), age (p = 0.01), and adjuvant endocrine treatment duration (p < 0.001). CONCLUSIONS: Qualitative evaluation of BPE, FGT, and mammographic MD changes cannot predict which patients are more likely to benefit from adjuvant endocrine treatment.

A machine learning model that classifies breast cancer pathologic complete response on MRI post-neoadjuvant chemotherapy.

BACKGROUND: For breast cancer patients undergoing neoadjuvant chemotherapy (NAC), pathologic complete response (pCR; no invasive or in situ) cannot be assessed non-invasively so all patients undergo surgery. The aim of our study was to develop and validate a radiomics classifier that classifies breast cancer pCR post-NAC on MRI prior to surgery. METHODS: This retrospective study included women treated with NAC for breast cancer from 2014 to 2016 with (1) pre- and post-NAC breast MRI and (2) post-NAC surgical pathology report assessing response. Automated radiomics analysis of pre- and post-NAC breast MRI involved image segmentation, radiomics feature extraction, feature pre-filtering, and classifier building through recursive feature elimination random forest (RFE-RF) machine learning. The RFE-RF classifier was trained with nested five-fold cross-validation using (a) radiomics only (model 1) and (b) radiomics and molecular subtype (model 2). Class imbalance was addressed using the synthetic minority oversampling technique. RESULTS: Two hundred seventy-three women with 278 invasive breast cancers were included; the training set consisted of 222 cancers (61 pCR, 161 no-pCR; mean age 51.8 years, SD 11.8), and the independent test set consisted of 56 cancers (13 pCR, 43 no-pCR; mean age 51.3 years, SD 11.8). There was no significant difference in pCR or molecular subtype between the training and test sets. Model 1 achieved a cross-validation AUROC of 0.72 (95% CI 0.64, 0.79) and a similarly accurate (P = 0.1) AUROC of 0.83 (95% CI 0.71, 0.94) in both the training and test sets. Model 2 achieved a cross-validation AUROC of 0.80 (95% CI 0.72, 0.87) and a similar (P = 0.9) AUROC of 0.78 (95% CI 0.62, 0.94) in both the training and test sets. CONCLUSIONS: This study validated a radiomics classifier combining radiomics with molecular subtypes that accurately classifies pCR on MRI post-NAC.

Ultrafast dynamic contrast-enhanced breast MRI may generate prognostic imaging markers of breast cancer.

BACKGROUND: Ultrafast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI)-derived kinetic parameters have demonstrated at least equivalent accuracy to standard DCE-MRI in differentiating malignant from benign breast lesions. However, it is unclear if they have any efficacy as prognostic imaging markers. The aim of this study was to investigate the relationship between ultrafast DCE-MRI-derived kinetic parameters and breast cancer characteristics. METHODS: Consecutive breast MRI examinations between February 2017 and January 2018 were retrospectively reviewed to determine those examinations that meet the following inclusion criteria: (1) BI-RADS 4-6 MRI performed on a 3T scanner with a 16-channel breast coil and (2) a hybrid clinical protocol with 15 phases of ultrafast DCE-MRI (temporal resolution of 2.7-4.6 s) followed by early and delayed phases of standard DCE-MRI. The study included 125 examinations with 142 biopsy-proven breast cancer lesions. Ultrafast DCE-MRI-derived kinetic parameters (maximum slope [MS] and bolus arrival time [BAT]) were calculated for the entire volume of each lesion. Comparisons of these parameters between different cancer characteristics were made using generalized estimating equations, accounting for the presence of multiple lesions per patient. All comparisons were exploratory and adjustment for multiple comparisons was not performed; P values < 0.05 were considered statistically significant. RESULTS: Significantly larger MS and shorter BAT were observed for invasive carcinoma than ductal carcinoma in situ (DCIS) (P < 0.001 and P = 0.008, respectively). Significantly shorter BAT was observed for invasive carcinomas with more aggressive characteristics than those with less aggressive characteristics: grade 3 vs. grades 1-2 (P = 0.025), invasive ductal carcinoma vs. invasive lobular carcinoma (P = 0.002), and triple negative or HER2 type vs. luminal type (P < 0.001). CONCLUSIONS: Ultrafast DCE-MRI-derived parameters showed a strong relationship with some breast cancer characteristics, especially histopathology and molecular subtype.

Improved characterization of sub-centimeter enhancing breast masses on MRI with radiomics and machine learning in BRCA mutation carriers.

OBJECTIVES: To investigate whether radiomics features extracted from MRI of BRCA-positive patients with sub-centimeter breast masses can be coupled with machine learning to differentiate benign from malignant lesions using model-free parameter maps. METHODS: In this retrospective study, BRCA-positive patients who had an MRI from November 2013 to February 2019 that led to a biopsy (BI-RADS 4) or imaging follow-up (BI-RADS 3) for sub-centimeter lesions were included. Two radiologists assessed all lesions independently and in consensus according to BI-RADS. Radiomics features were calculated using open-source CERR software. Univariate analysis and multivariate modeling were performed to identify significant radiomics features and clinical factors to be included in a machine learning model to differentiate malignant from benign lesions. RESULTS: Ninety-six BRCA mutation carriers (mean age at biopsy = 45.5 ± 13.5 years) were included. Consensus BI-RADS classification assessment achieved a diagnostic accuracy of 53.4%, sensitivity of 75% (30/40), specificity of 42.1% (32/76), PPV of 40.5% (30/74), and NPV of 76.2% (32/42). The machine learning model combining five parameters (age, lesion location, GLCM-based correlation from the pre-contrast phase, first-order coefficient of variation from the 1st post-contrast phase, and SZM-based gray level variance from the 1st post-contrast phase) achieved a diagnostic accuracy of 81.5%, sensitivity of 63.2% (24/38), specificity of 91.4% (64/70), PPV of 80.0% (24/30), and NPV of 82.1% (64/78). CONCLUSIONS: Radiomics analysis coupled with machine learning improves the diagnostic accuracy of MRI in characterizing sub-centimeter breast masses as benign or malignant compared with qualitative morphological assessment with BI-RADS classification alone in BRCA mutation carriers. KEY POINTS: • Radiomics and machine learning can help differentiate benign from malignant breast masses even if the masses are small and morphological features are benign. • Radiomics and machine learning analysis showed improved diagnostic accuracy, specificity, PPV, and NPV compared with qualitative morphological assessment alone.

Differentiation between subcentimeter carcinomas and benign lesions using kinetic parameters derived from ultrafast dynamic contrast-enhanced breast MRI.

OBJECTIVES: This study aims to evaluate ultrafast DCE-MRI-derived kinetic parameters that reflect contrast agent inflow effects in differentiating between subcentimeter BI-RADS 4-5 breast carcinomas and benign lesions. METHODS: We retrospectively reviewed consecutive 3-T MRI performed from February to October 2017, during which ultrafast DCE-MRI was performed as part of a hybrid clinical protocol with conventional DCE-MRI. In total, 301 female patients with 369 biopsy-proven breast lesions were included. Ultrafast DCE-MRI was acquired continuously over approximately 60 s (temporal resolution, 2.7-7.1 s/phase) starting simultaneously with the start of contrast injection. Four ultrafast DCE-MRI-derived kinetic parameters (maximum slope [MS], contrast enhancement ratio [CER], bolus arrival time [BAT], and initial area under gadolinium contrast agent concentration [IAUGC]) and one conventional DCE-MRI-derived kinetic parameter (signal enhancement ratio [SER]) were calculated for each lesion. Wilcoxon rank sum test or Fisher's exact test was performed to compare kinetic parameters, volume, diameter, age, and BI-RADS morphological descriptors between subcentimeter carcinomas and benign lesions. Univariate/multivariate logistic regression analyses were performed to determine predictive parameters for subcentimeter carcinomas. RESULTS: In total, 125 lesions (26 carcinomas and 99 benign lesions) were identified as BI-RADS 4-5 subcentimeter lesions. Subcentimeter carcinomas demonstrated significantly larger MS and SER and shorter BAT than benign lesions (p = 0.0117, 0.0046, and 0.0102, respectively). MS, BAT, and age were determined as significantly predictive for subcentimeter carcinoma (p = 0.0208, 0.0023, and < 0.0001, respectively). CONCLUSIONS: Ultrafast DCE-MRI-derived kinetic parameters may be useful in differentiating subcentimeter BI-RADS 4 and 5 carcinomas from benign lesions. KEY POINTS: • Ultrafast DCE-MRI can generate kinetic parameters, effectively differentiating breast carcinomas from benign lesions. • Subcentimeter carcinomas demonstrated significantly larger maximum slope and shorter bolus arrival time than benign lesions. • Maximum slope and bolus arrival time contribute to better management of suspicious subcentimeter breast lesions.

Radiomic signatures with contrast-enhanced magnetic resonance imaging for the assessment of breast cancer receptor status and molecular subtypes: initial results.

BACKGROUND: To evaluate the diagnostic performance of radiomic signatures extracted from contrast-enhanced magnetic resonance imaging (CE-MRI) for the assessment of breast cancer receptor status and molecular subtypes. METHODS: One hundred and forty-three patients with biopsy-proven breast cancer who underwent CE-MRI at 3 T were included in this IRB-approved HIPAA-compliant retrospective study. The training dataset comprised 91 patients (luminal A, n = 49; luminal B, n = 8; HER2-enriched, n = 11; triple negative, n = 23), while the validation dataset comprised 52 patients from a second institution (luminal A, n = 17; luminal B, n = 17; triple negative, n = 18). Radiomic analysis of manually segmented tumors included calculation of features derived from the first-order histogram (HIS), co-occurrence matrix (COM), run-length matrix (RLM), absolute gradient (GRA), autoregressive model (ARM), discrete Haar wavelet transform (WAV), and lesion geometry (GEO). Fisher, probability of error and average correlation (POE + ACC), and mutual information coefficients were used for feature selection. Linear discriminant analysis followed by k-nearest neighbor classification (with leave-one-out cross-validation) was used for pairwise radiomic-based separation of receptor status and molecular subtypes. Histopathology served as the standard of reference. RESULTS: In the training dataset, radiomic signatures yielded the following accuracies > 80%: luminal B vs. luminal A, 84.2% (mainly based on COM features); luminal B vs. triple negative, 83.9% (mainly based on GEO features); luminal B vs. all others, 89% (mainly based on COM features); and HER2-enriched vs. all others, 81.3% (mainly based on COM features). Radiomic signatures were successfully validated in the separate validation dataset for luminal A vs. luminal B (79.4%) and luminal B vs. triple negative (77.1%). CONCLUSIONS: In this preliminary study, radiomic signatures with CE-MRI enable the assessment of breast cancer receptor status and molecular subtypes with high diagnostic accuracy. These results need to be confirmed in future larger studies.

Limited role of DWI with apparent diffusion coefficient mapping in breast lesions presenting as non-mass enhancement on dynamic contrast-enhanced MRI.

BACKGROUND: Available data proving the value of DWI for breast cancer diagnosis is mainly for enhancing masses; DWI may be less sensitive and specific in non-mass enhancement (NME) lesions. The objective of this study was to assess the diagnostic accuracy of DWI using different ROI measurement approaches and ADC metrics in breast lesions presenting as NME lesions on dynamic contrast-enhanced (DCE) MRI. METHODS: In this retrospective study, 95 patients who underwent multiparametric MRI with DCE and DWI from September 2007 to July 2013 and who were diagnosed with a suspicious NME (BI-RADS 4/5) were included. Twenty-nine patients were excluded for lesion non-visibility on DWI (n = 24: 12 benign and 12 malignant) and poor DWI quality (n = 5: 1 benign and 4 malignant). Two readers independently assessed DWI and DCE-MRI findings in two separate randomized readings using different ADC metrics and ROI approaches. NME lesions were classified as either benign (> 1.3 × 10-3 mm2/s) or malignant (≤ 1.3 × 10-3 mm2/s). Histopathology was the standard of reference. ROC curves were plotted, and AUCs were determined. Concordance correlation coefficient (CCC) was measured. RESULTS: There were 39 malignant (59%) and 27 benign (41%) lesions in 66 (65 women, 1 man) patients (mean age, 51.8 years). The mean ADC value of the darkest part of the tumor (Dptu) achieved the highest diagnostic accuracy, with AUCs of up to 0.71. Inter-reader agreement was highest with Dptu ADC max (CCC 0.42) and lowest with the point tumor (Ptu) ADC min (CCC = - 0.01). Intra-reader agreement was highest with Wtu ADC mean (CCC = 0.44 for reader 1, 0.41 for reader 2), but this was not associated with the highest diagnostic accuracy. CONCLUSIONS: Diagnostic accuracy of DWI with ADC mapping is limited in NME lesions. Thirty-one percent of lesions presenting as NME on DCE-MRI could not be evaluated with DWI, and therefore, DCE-MRI remains indispensable. Best results were achieved using Dptu 2D ROI measurement and ADC mean.

A multiparametric [18F]FDG PET/MRI diagnostic model including imaging biomarkers of the tumor and contralateral healthy breast tissue aids breast cancer diagnosis.

PURPOSE: To develop a multiparametric [18F]FDG positron emission tomography/magnetic resonance imaging (PET/MRI) model for breast cancer diagnosis incorporating imaging biomarkers of breast tumors and contralateral healthy breast tissue. METHODS: In this prospective study and retrospective data analysis, 141 patients (mean 57 years) with an imaging abnormality detected on mammography and/or ultrasound (BI-RADS 4/5) underwent combined multiparametric [18F]FDG PET/MRI with PET/computed tomography and multiparametric MRI of the breast at 3 T. Images were evaluated and the following were recorded: for the tumor, BI-RADS descriptors on dynamic contrast-enhanced (DCE)-MRI, mean apparent diffusion co-efficient (ADCmean) on diffusion-weighted imaging (DWI), and maximum standard uptake value (SUVmax) on [18F]FDG-PET; and for the contralateral healthy breast, background parenchymal enhancement (BPE) and amount of fibroglandular tissue (FGT) on DCE-MRI, ADCmean on DWI, and SUVmax. Histopathology served as standard of reference. Uni-, bi-, and multivariate logistic regression analyses were performed to assess the relationships between malignancy and imaging features. Predictive discrimination of benign and malignant breast lesions was examined using area under the receiver operating characteristic curve (AUC). RESULTS: There were 100 malignant and 41 benign lesions (size: median 1.9, range 0.5-10 cm). The multivariate regression model incorporating significant univariate predictors identified tumor enhancement kinetics (P = 0.0003), tumor ADCmean (P < 0.001), and BPE of the contralateral healthy breast (P = 0.0019) as independent predictors for breast cancer diagnosis. Other biomarkers did not reach significance. Combination of the three significant biomarkers achieved an AUC value of 0.98 for breast cancer diagnosis. CONCLUSION: A multiparametric [18F]FDG PET/MRI diagnostic model incorporating both qualitative and quantitative parameters of the tumor and the healthy contralateral tissue aids breast cancer diagnosis.

Characterization of Sub-1 cm Breast Lesions Using Radiomics Analysis.

BACKGROUND: Small breast lesions are difficult to visually categorize due to the inherent lack of morphological and kinetic detail. PURPOSE: To assess the efficacy of radiomics analysis in discriminating small benign and malignant lesions utilizing model free parameter maps. STUDY TYPE: Retrospective, single center. POPULATION: In all, 149 patients, with a total of 165 lesions scored as BI-RADS 4 or 5 on MRI, with an enhancing volume of <0.52 cm3 . FIELD STRENGTH/SEQUENCE: Higher spatial resolution T1 -weighted dynamic contrast-enhanced imaging with a temporal resolution of ~90 seconds performed at 3.0T. ASSESSMENT: Parameter maps reflecting initial enhancement, overall enhancement, area under the enhancement curve, and washout were generated. Heterogeneity measures based on first-order statistics, gray level co-occurrence matrices, run length matrices, size zone matrices, and neighborhood gray tone difference matrices were calculated. Data were split into a training dataset (~75% of cases) and a test dataset (~25% of cases). STATISTICAL TESTS: Comparison of medians was assessed using the nonparametric Mann-Whitney U-test. The Spearman rank correlation coefficient was utilized to determine significant correlations between individual features. Finally, a support vector machine was employed to build multiparametric predictive models. RESULTS: Univariate analysis revealed significant differences between benign and malignant lesions for 58/133 calculated features (P < 0.05). Support vector machine analysis resulted in areas under the curve (AUCs) ranging from 0.75-0.81. High negative (>89%) and positive predictive values (>83%) were found for all models. DATA CONCLUSION: Radiomics analysis of small contrast-enhancing breast lesions is of value. Texture features calculated from later timepoints on the enhancement curve appear to offer limited additional value when compared with features determined from initial enhancement for this patient cohort. LEVEL OF EVIDENCE: 4 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2019;50:1468-1477.

MRI evaluation of axillary and intramammary lymph nodes in the postoperative period.

Our study aimed to evaluate if breast-conserving surgery and adjuvant treatment could affect the morphological features of axillary and intramammary lymph nodes on magnetic resonance imaging (MRI) in patients with invasive breast cancer and clinically negative axilla. In this single-center study, we retrospectively evaluated 50 patients who had (a) breast-conserving surgery, (b) clinically negative axilla, (c) preoperative MRI within 3 months before surgery, and (d) postoperative MRI within 12 months after surgery. Axillary and intramammary lymph nodes on postoperative MRI were identified and then compared with preoperative MRI by two breast radiologists with regards to the following: enlargement, cortical thickening, presence of fatty hilum, irregularity, heterogeneity, matting, and axillary lymph node asymmetry. Three hundred and two axillary and eight intramammary lymph nodes were evaluated. Enlargement and cortical thickening were seen in 5/50 (10%) patients in three axillary and two intramammary lymph nodes. None of the lymph nodes on postoperative MRI demonstrated occurrence of lack of fatty hilum, irregularity, heterogeneity, matting or axillary lymph node asymmetry. No evidence of recurrence was observed on 2-year follow-up. Lymph node enlargement and cortical thickening may be observed in a few patients in the postoperative period. Nevertheless, in patients with clinically negative axilla, these changes in morphology are often related to treatment rather than malignancy and favor short-term follow-up as an alternative to lymph node biopsy.

[WITHDRAWN] Predicting breast cancer with AI for individual risk-adjusted MRI screening and early detection.

This paper has been withdrawn by Lukas Hirsch. Major revisions and rewriting in progress.

Early Detection of Breast Cancer in MRI Using AI.

RATIONALE AND OBJECTIVES: To develop and evaluate an AI algorithm that detects breast cancer in MRI scans up to one year before radiologists typically identify it, potentially enhancing early detection in high-risk women. MATERIALS AND METHODS: A convolutional neural network (CNN) AI model, pre-trained on breast MRI data, was fine-tuned using a retrospective dataset of 3029 MRI scans from 910 patients. These contained 115 cancers that were diagnosed within one year of a negative MRI. The model aimed to identify these cancers, with the goal of predicting cancer development up to one year in advance. The network was fine-tuned and tested with 10-fold cross-validation. Mean age of patients was 52 years (range, 18-88 years), with average follow-up of 4.3 years (range 1-12 years). RESULTS: The AI detected cancers one year earlier with an area under the ROC curve of 0.72 (0.67-0.76). Retrospective analysis by a radiologist of the top 10% highest risk MRIs as ranked by the AI could have increased early detection by up to 30%. (35/115, CI:22.2-39.7%, 30% sensitivity). A radiologist identified a visual correlate to biopsy-proven cancers in 83 of prior-year MRIs (83/115, CI: 62.1-79.4%). The AI algorithm identified the anatomic region where cancer would be detected in 66 cases (66/115, CI:47.8-66.5%); with both agreeing in 54 cases (54/115, CI:%37.5-56.4%). CONCLUSION: This novel AI-aided re-evaluation of "benign" breasts shows promise for improving early breast cancer detection with MRI. As datasets grow and image quality improves, this approach is expected to become even more impactful.

Preoperative localization of breast lesions: Comparing digital breast tomosynthesis-guided radioactive seed localization versus standard 2D stereotactic radioactive seed localization.

PURPOSE: To compare single seed digital breast tomosynthesis-guided radioseed localization (DBT-L) to standard 2D stereotactic-guided radioseed localization (SGL) of the breast. METHODS: A retrospective review of a large tertiary cancer center's database yielded 68 women who underwent preoperative DBT-L from March 2019-December 2019 and a matched cohort of 65 women who underwent SGL during the same period. The electronic medical record and radiology were reviewed for patient characteristics including breast density, exam technique, pre- and post-operative pathology, exam duration, and radiation dose to the patient. To compare margin outcomes between the groups, the chi-square test of independence was used; to compare continuous outcomes such as exam duration and total dose, the Wilcoxon rank sum test was used. RESULTS: DBT-L and SGL localization targets included biopsy marker (62/68, 91% vs 55/65, 85%), distortion (4/68, 6% vs 2/65, <3%), focal asymmetry (1/68 and 1/65, < 2% for both), calcifications (1/68, <2% vs 4/65, 6%), and anatomic landmarks (0% vs 3/65, 5%). 72% and 71% of localizations were performed for malignant pathology in the DBT-L and SGL groups, respectively. The median duration of DBT-L was 8.3 min vs 10.3 min for SGL, representing statistically significant time savings (p = 0.003). The median total organ dose of DBT-L was 8.6 mGy vs 10.4 mGy for SGL, representing statistically significant dose savings (p = 0.018). The incidence of positive margins at surgery was not statistically different between the groups (p = 0.26). CONCLUSION: DBT-L demonstrates both time and dose savings for the patient compared to SGL without compromising surgical outcome.

Automated Triage of Screening Breast MRI Examinations in High-Risk Women Using an Ensemble Deep Learning Model.

OBJECTIVES: The aim of the study is to develop and evaluate the performance of a deep learning (DL) model to triage breast magnetic resonance imaging (MRI) findings in high-risk patients without missing any cancers. MATERIALS AND METHODS: In this retrospective study, 16,535 consecutive contrast-enhanced MRIs performed in 8354 women from January 2013 to January 2019 were collected. From 3 New York imaging sites, 14,768 MRIs were used for the training and validation data set, and 80 randomly selected MRIs were used for a reader study test data set. From 3 New Jersey imaging sites, 1687 MRIs (1441 screening MRIs and 246 MRIs performed in recently diagnosed breast cancer patients) were used for an external validation data set. The DL model was trained to classify maximum intensity projection images as "extremely low suspicion" or "possibly suspicious." Deep learning model evaluation (workload reduction, sensitivity, specificity) was performed on the external validation data set, using a histopathology reference standard. A reader study was performed to compare DL model performance to fellowship-trained breast imaging radiologists. RESULTS: In the external validation data set, the DL model triaged 159/1441 of screening MRIs as "extremely low suspicion" without missing a single cancer, yielding a workload reduction of 11%, a specificity of 11.5%, and a sensitivity of 100%. The model correctly triaged 246/246 (100% sensitivity) of MRIs in recently diagnosed patients as "possibly suspicious." In the reader study, 2 readers classified MRIs with a specificity of 93.62% and 91.49%, respectively, and missed 0 and 1 cancer, respectively. On the other hand, the DL model classified MRIs with a specificity of 19.15% and missed 0 cancers, highlighting its potential use not as an independent reader but as a triage tool. CONCLUSIONS: Our automated DL model triages a subset of screening breast MRIs as "extremely low suspicion" without misclassifying any cancer cases. This tool may be used to reduce workload in standalone mode, to shunt low suspicion cases to designated radiologists or to the end of the workday, or to serve as base model for other downstream AI tools.

Radiologist-Level Performance by Using Deep Learning for Segmentation of Breast Cancers on MRI Scans.

PURPOSE: To develop a deep network architecture that would achieve fully automated radiologist-level segmentation of cancers at breast MRI. MATERIALS AND METHODS: In this retrospective study, 38 229 examinations (composed of 64 063 individual breast scans from 14 475 patients) were performed in female patients (age range, 12-94 years; mean age, 52 years ± 10 [standard deviation]) who presented between 2002 and 2014 at a single clinical site. A total of 2555 breast cancers were selected that had been segmented on two-dimensional (2D) images by radiologists, as well as 60 108 benign breasts that served as examples of noncancerous tissue; all these were used for model training. For testing, an additional 250 breast cancers were segmented independently on 2D images by four radiologists. Authors selected among several three-dimensional (3D) deep convolutional neural network architectures, input modalities, and harmonization methods. The outcome measure was the Dice score for 2D segmentation, which was compared between the network and radiologists by using the Wilcoxon signed rank test and the two one-sided test procedure. RESULTS: The highest-performing network on the training set was a 3D U-Net with dynamic contrast-enhanced MRI as input and with intensity normalized for each examination. In the test set, the median Dice score of this network was 0.77 (interquartile range, 0.26). The performance of the network was equivalent to that of the radiologists (two one-sided test procedures with radiologist performance of 0.69-0.84 as equivalence bounds, P < .001 for both; n = 250). CONCLUSION: When trained on a sufficiently large dataset, the developed 3D U-Net performed as well as fellowship-trained radiologists in detailed 2D segmentation of breast cancers at routine clinical MRI.Keywords: MRI, Breast, Segmentation, Supervised Learning, Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning AlgorithmsPublished under a CC BY 4.0 license. Supplemental material is available for this article.

Assessing PD-L1 Expression Status Using Radiomic Features from Contrast-Enhanced Breast MRI in Breast Cancer Patients: Initial Results.

The purpose of this retrospective study was to assess whether radiomics analysis coupled with machine learning (ML) based on standard-of-care dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) can predict PD-L1 expression status in patients with triple negative breast cancer, and to compare the performance of this approach with radiologist review. Patients with biopsy-proven triple negative breast cancer who underwent pre-treatment breast MRI and whose PD-L1 status was available were included. Following 3D tumor segmentation and extraction of radiomic features, radiomic features with significant differences between PD-L1+ and PD-L1- patients were determined, and a final predictive model to predict PD-L1 status was developed using a coarse decision tree and five-fold cross-validation. Separately, all lesions were qualitatively assessed by two radiologists independently according to the BI-RADS lexicon. Of 62 women (mean age 47, range 31-81), 27 had PD-L1- tumors and 35 had PD-L1+ tumors. The final radiomics model to predict PD-L1 status utilized three MRI parameters, i.e., variance (FO), run length variance (RLM), and large zone low grey level emphasis (LZLGLE), for a sensitivity of 90.7%, specificity of 85.1%, and diagnostic accuracy of 88.2%. There were no significant associations between qualitative assessed DCE-MRI imaging features and PD-L1 status. Thus, radiomics analysis coupled with ML based on standard-of-care DCE-MRI is a promising approach to derive prognostic and predictive information and to select patients who could benefit from anti-PD-1/PD-L1 treatment.