Multiparametric MRI model with dynamic contrast-enhanced and diffusion-weighted imaging enables breast cancer diagnosis with high accuracy

Quantitative Assessment of Breast Tumor: Comparison of Four Methods of Positioning Region of Interest for Synthetic Relaxometry and Diffusion Measurement

RATIONALE AND OBJECTIVES

To compare the differences in apparent diffusion coefficient (ADC) and synthetic magnetic resonance (MR) measurements of four region of interest (ROI) placement methods for breast tumor and to investigate their diagnostic performance.

METHODS

110 (70 malignant, 40 benign) newly diagnosed breast tumors were evaluated. The patients underwent 3.0 T MR examinations including diffusion-weighted imaging and synthetic MR. Two radiologists independently measured ADCs, T1 relaxation time (T1), T2 relaxation time (T2), and proton density (PD) using four ROI methods: round, square, freehand, and whole-tumor volume (WTV). The interclass correlation coefficient (ICC) was used to assess their measurement reliability. Diagnostic performance was evaluated using multivariate logistic regression analysis and the receiver operating characteristic (ROC) curves.

RESULTS

The mean values of all ROI methods showed good or excellent interobserver reproducibility (0.79-0.99) and showed the best diagnostic performance compared to the minimum and maximum values. The square ROI exhibited superior performance in differentiating between benign from malignant breast lesions, followed by the freehand ROI. T2, PD, and ADC values were significantly lower in malignant breast lesions compared to benign ones for all ROI methods (p < 0.05). Multiparameters of T2 + ADC demonstrated the highest AUC values (0.82-0.95), surpassing the diagnostic efficacy of ADC or T2 alone (p < 0.05).

CONCLUSION

ROI placement significantly influences ADC and synthetic MR values measured in breast tumors. Square ROI and mean values showed superior performance in differentiating benign and malignant breast lesions. The multiparameters of T2 + ADC surpassed the diagnostic efficacy of a single parameter.

Effect of contrast agent on T2-weighted fat-suppressed imaging and diffusion-weighted imaging in the diagnosis of breast tumors

Background

Although previous studies have shown that the injection of contrast agents can improve image quality, the specific impact of this on T2-weighted fat-suppressed (T2 FS) and diffusion-weighted imaging (DWI) sequences in the diagnosis of breast cancer remains incompletely understood. In particular, there is insufficient research on how contrast agents affect the signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and apparent diffusion coefficient (ADC) values within these sequences, and how these changes influence the diagnosis of benign and malignant breast tumors.

Methods

Breast magnetic resonance images (MRI) were obtained from 178 consecutive patients on a 3T scanner. The SNR and CNR of lesions on T2 FS sequence were calculated before and after contrast agent injection and compared. Differences between pre- and post-contrast ADC in identifying different tumor types were compared using the Kruskal-Wallis H-test and the paired comparison test. The accuracy of ADC values between pre- and post-contrast in distinguishing benign and malignant breast masses was assessed using receiver operating characteristic (ROC) curves.

Results

The SNR and CNR of T2 FS sequence increased after contrast injection, and especially for invasive cancer and benign tumor, the increase was significant. For DWI, there was a slight increase or decrease of ADC values after contrast injection, but the ADC values before and after contrast had a similar effect in identifying different types of tumors. In the ROC curve analysis for assessing benign and malignant breast tumors, the area under the curve (AUC) before and after contrast showed similar results.

Conclusions

Contrast agent injection can improve the SNR and CNR of T2 FS sequence, thus providing higher quality images for the diagnosis of breast lesions. Furthermore, injection of contrast agent had little effect on the ability of ADC values to identify different types of lesions and both ADC values before and after the contrast agent were able to distinguish between benign and malignant tumors with almost the same accuracy.

How to manage type 2 curve dilemma in dynamic contrast-enhanced magnetic resonance imaging of the breast: diffusion-weighted imaging or early phase enhancement kinetics?

BACKGROUND

Type 2 time-intensity curves can indicate both malignant and benign breast lesions in dynamic contrast enhanced magnetic resonance imaging (DCE-MRI).

PURPOSE

To investigate whether diffusion-weighted imaging (DWI) or early phase kinetics of DCE-MRI is practical to discriminate breast masses that depict type 2 curve in DCE-MRI.

MATERIAL AND METHODS

We retrospectively included 107 lesions in 97 patients with type 2 curves in DCE-MRI. Morphological characteristics, early phase dynamic parameters on DCE-MRI, and apparent diffusion coefficient (ADC) values on DWI were evaluated. Diagnostic thresholds of ADC and early phase maximum enhancement ratio (EPMER) to distinguish between benign and malignant masses were calculated. Strongest predictors of malignancy were determined to build the most effective diagnostic model.

RESULTS

DWI, EPMER, and all morphological features were found statistically significant to discriminate malignancy (P <0.05). The thresholds of ADC and EPMER were assigned as 1.0 ×10-3 mm2/s and 72%, respectively. The sensitivity and specificity were 80% and 97% for ADC, and 93% and 60% for EPMER, respectively. Two models were established. Model 1 comprised ADC and the lesion margin. Model 2 consisted of ADC, margin, and EPMER with a high specificity (99%) and positive predictive value (97%).

CONCLUSION

When combined with DWI, early phase wash-in data provide diagnostic improvement of breast masses presenting type 2 curve in the late phase of DCE-MRI, especially for specificity. Future studies are required to support our findings for the need of a cross-validation.

Improved Differential Diagnosis Based on BI-RADS Descriptors and Apparent Diffusion Coefficient for Breast Lesions: A Multiparametric MRI Analysis as Compared to Kaiser Score

RATIONALE AND OBJECTIVES

To develop the nomogram utilizing the American College of Radiology BI-RADS descriptors, clinical features, and apparent diffusion coefficient (ADC) to differentiate benign from malignant breast lesions.

MATERIALS AND METHODS

A total of 341 lesions (161 malignant and 180 benign) were included. Clinical data and imaging features were reviewed. Univariable and multivariable logistic regression analyses were performed to determine the independent variables. ADC as a continuous or classified into binary form with a cutoff value of 1.30 × 10-3 mm2/s, incorporated other independent predictors to construct two nomograms, respectively. Receiver operating curve and calibration plot was employed to test the models' discriminative ability. The diagnostic performance between the developed model and the Kaiser score (KS) was also compared.

RESULTS

In both models, high patient age, the presence of root sign, time-intensity curves (TICs) types (plateau and washout), heterogenous internal enhancement, the presence of peritumoral edema, and ADC were independently associated with malignancy. The AUCs of two multivariable models (AUC, 0.957; 95% CI: 0.929-0.976 and AUC, 0.958; 95% CI: 0.931-0.976) were significantly higher than that of the KS (AUC, 0.919, 95% CI: 0.885-0.946; both P < 0.001). At the same sensitivity of 95.7%, our models showed an increase in specificity by 5.56% (P = 0.076) and 6.11% (P = 0.035), respectively, as compared to the KS.

CONCLUSION

The models incorporating MRI features (root sign, TIC, margins, internal enhancement, and presence of edema), quantitative ADC value, and patient age showed improved diagnostic performance and might have avoided more unnecessary biopsies in comparison with the KS, although further external validation is required.

Ensemble Learning for Breast Cancer Lesion Classification: A Pilot Validation Using Correlated Spectroscopic Imaging and Diffusion-Weighted Imaging

The main objective of this work was to evaluate the application of individual and ensemble machine learning models to classify malignant and benign breast masses using features from two-dimensional (2D) correlated spectroscopy spectra extracted from five-dimensional echo-planar correlated spectroscopic imaging (5D EP-COSI) and diffusion-weighted imaging (DWI). Twenty-four different metabolite and lipid ratios with respect to diagonal fat peaks (1.4 ppm, 5.4 ppm) from 2D spectra, and water and fat peaks (4.7 ppm, 1.4 ppm) from one-dimensional non-water-suppressed (NWS) spectra were used as the features. Additionally, water fraction, fat fraction and water-to-fat ratios from NWS spectra and apparent diffusion coefficients (ADC) from DWI were included. The nine most important features were identified using recursive feature elimination, sequential forward selection and correlation analysis. XGBoost (AUC: 93.0%, Accuracy: 85.7%, F1-score: 88.9%, Precision: 88.2%, Sensitivity: 90.4%, Specificity: 84.6%) and GradientBoost (AUC: 94.3%, Accuracy: 89.3%, F1-score: 90.7%, Precision: 87.9%, Sensitivity: 94.2%, Specificity: 83.4%) were the best-performing models. Conventional biomarkers like choline, myo-Inositol, and glycine were statistically significant predictors. Key features contributing to the classification were ADC, 2D diagonal peaks at 0.9 ppm, 2.1 ppm, 3.5 ppm, and 5.4 ppm, cross peaks between 1.4 and 0.9 ppm, 4.3 and 4.1 ppm, 2.3 and 1.6 ppm, and the triglyceryl-fat cross peak. The results highlight the contribution of the 2D spectral peaks to the model, and they demonstrate the potential of 5D EP-COSI for early breast cancer detection.

Diffusion-Weighted Imaging as a Quantitative Imaging Biomarker for Predicting Proliferation Rate in Hepatocellular Carcinoma: Developing a Radiomics Nomogram

PURPOSE

This study aimed to explore the predictive performance of diffusion-weighted imaging with apparent diffusion coefficient map in predicting the proliferation rate of hepatocellular carcinoma and to develop a radiomics-based nomogram.

METHODS

This was a single-center retrospective study. A total of 110 patients were enrolled. The sample included 38 patients with low Ki67 expression (Ki67 ≤10%) and 72 with high Ki67 expression (Ki67 >10%) as demonstrated by surgical pathology. Patients were randomly divided into either a training (n = 77) or validation (n = 33) cohort. Diffusion-weighted imaging with apparent diffusion coefficient maps was used to extract radiomic features and the signal intensity values of tumor (SItumor), normal liver (SIliver), and background noise (SIbackground) from all samples. Subsequently, the clinical model, radiomic model, and fusion model (with clinical data and radiomic signature) were developed and validated.

RESULTS

The area under the curve (AUC) of the clinical model for predicting the Ki67 expression including serum α-fetoprotein level (P = 0.010), age (P = 0.015), and signal noise ratio (P = 0.026) was 0.799 and 0.715 in training and validation cohorts, respectively. The AUC of the radiomic model constructed by 9 selected radiomic features was 0.833 and 0.772 in training and validation cohorts, respectively. The AUC of the fusion model containing serum α-fetoprotein level (P = 0.011), age (P = 0.019), and rad score (P < 0.001) was 0.901 and 0.781 in training and validation cohorts, respectively.

CONCLUSIONS

Diffusion-weighted imaging as a quantitative imaging biomarker can predict Ki67 expression level in hepatocellular carcinoma across various models.

Diffusion-weighted MRI of total hip arthroplasty for classification of synovial reactions: A pilot study

BACKGROUND

Conventional quantitative diffusion-weighted imaging (DWI) is sensitive to changes in tissue microstructure, but its application to evaluating patients with orthopaedic hardware has generally been limited due to metallic susceptibility artifacts. The apparent diffusion coefficient (ADC) and T2-values from a multi-spectral imaging (MSI) DWI combined with 2D multi-spectral imaging with a 2D periodically rotated overlapping parallel lines with enhanced reconstruction (2D-MSI PROPELLER DWI) based sequence and a MAVRIC based T2 mapping sequence, respectively, may mitigate the artifact and provide additional quantitative information on synovial reactions in individuals with total hip arthroplasty (THA). The aim of this pilot study is to utilize a 2D-MSI PROPELLER DWI and a MAVRIC-based T2 mapping to evaluate ADC and T2-values of synovial reactions in patients with THA.

METHODS

Coronal morphologic MRIs from THA patients underwent evaluation of the synovium and were assigned a synovial classification of 'normal', or 'grouped abnormal' (consisting of sub-groups 'infection', 'polymeric', 'metallosis', 'adverse local tissue reaction' [ALTR], or 'non-specific') and type of synovial reaction present (fluid-like, solid-like, or mixed). Regions of interest (ROIs) were placed in synovial reactions for measurement of ADC and T2-values, obtained from the 2D-MSI PROPELLER DWI and T2-MAVRIC sequences, respectively. A one-way analysis of variance (ANOVA) and Kruskal-Wallis rank sum tests were used to compare the differences in ADC and T2-values across the different synovial reaction classifications. A Kruskal-Wallis test was used to compare the ROI areas for the ADC and T2-values. A principal component analysis (PCA) was performed to evaluate the possible effects of ADC values, size of the ADC ROI, T2-values, and size of the T2 ROI with respect to synovial reaction classification.

RESULTS

Differences of ADC and T2 among the individual synovial reactions were not found. A difference of ADC between 'normal' and 'grouped abnormal' synovial reactions was also not detected even as the ADC area of 'grouped abnormal' synovial reactions were significantly larger (p = 0.02). The 'grouped abnormal' synovial reactions had significantly shorter T2-values than 'normal' synovial reactions (p = 0.02), and that the T2 area of 'grouped abnormal' synovial reactions were significantly larger (p = 0.01). A larger ROI area on the T2-maps was observed in the mixed synovial reaction type as compared to the fluid-like reaction type area (p = 0.01). Heterogeneity was noted in calculated ADC and T2 maps. PCA analysis revealed obvious clustering by the 'normal' and 'grouped abnormal' classifications.

CONCLUSIONS

2D-MSI PROPELLER DWI and MAVRIC-T2 generate quantitative images of periprosthetic tissues within clinically feasible scan times. The combination of derived ADC and T2-values with area of synovial reaction may aid in differentiating normal from abnormal synovial reactions between types of synovial reactions in patients with THA.

Advanced Magnetic Resonance Imaging Modalities for Breast Cancer Diagnosis: An Overview of Recent Findings and Perspectives

Breast cancer is the most prevalent cancer among women and the leading cause of death. Diffusion-weighted imaging (DWI) and diffusion tensor imaging (DTI) are advanced magnetic resonance imaging (MRI) procedures that are widely used in the diagnostic and treatment evaluation of breast cancer. This review article describes the characteristics of new MRI methods and reviews recent findings on breast cancer diagnosis. This review study was performed on the literature sourced from scientific citation websites such as Google Scholar, PubMed, and Web of Science until July 2021. All relevant works published on the mentioned scientific citation websites were investigated. Because of the propensity of malignancies to limit diffusion, DWI can improve MRI diagnostic specificity. Diffusion tensor imaging gives additional information about diffusion directionality and anisotropy over traditional DWI. Recent findings showed that DWI and DTI and their characteristics may facilitate earlier and more accurate diagnosis, followed by better treatment. Overall, with the development of instruments and novel MRI modalities, it may be possible to diagnose breast cancer more effectively in the early stages.

Correlated MR spectroscopic imaging of breast cancer to investigate metabolites and lipids: acceleration and compressed sensing reconstruction

Objectives

The main objective of this work was to detect novel biomarkers in breast cancer by spreading the MR spectra over two dimensions in multiple spatial locations using an accelerated 5D EP-COSI technology.

Methods

The 5D EP-COSI data were non-uniformly undersampled with an acceleration factor of 8 and reconstructed using group sparsity-based compressed sensing reconstruction. Different metabolite and lipid ratios were then quantified and statistically analyzed for significance. Linear discriminant models based on the quantified metabolite and lipid ratios were generated. Spectroscopic images of the quantified metabolite and lipid ratios were also reconstructed.

Results

The 2D COSY spectra generated using the 5D EP-COSI technique showed differences among healthy, benign, and malignant tissues in terms of their mean values of metabolite and lipid ratios, especially the ratios of potential novel biomarkers based on unsaturated fatty acids, myo-inositol, and glycine. It is further shown the potential of choline and unsaturated lipid ratio maps, generated from the quantified COSY signals across multiple locations in the breast, to serve as complementary markers of malignancy that can be added to the multiparametric MR protocol. Discriminant models using metabolite and lipid ratios were found to be statistically significant for classifying benign and malignant tumor from healthy tissues.

Conclusions

Accelerated 5D EP-COSI technique demonstrates the potential to detect novel biomarkers such as glycine, myo-inositol, and unsaturated fatty acids in addition to commonly reported choline in breast cancer, and facilitates metabolite and lipid ratio maps which have the potential to play a significant role in breast cancer detection.

Advances in knowledge

This study presents the first evaluation of a multidimensional MR spectroscopic imaging technique for the detection of potentially novel biomarkers based on glycine, myo-inositol, and unsaturated fatty acids, in addition to commonly reported choline. Spatial mapping of choline and unsaturated fatty acid ratios with respect to water in malignant and benign breast masses are also shown. These metabolic characteristics may serve as additional biomarkers for improving the diagnostic and therapeutic evaluation of breast cancer.

Assessment of breast lesions by the Kaiser score for differential diagnosis on MRI: the added value of ADC and machine learning modeling

OBJECTIVES

To evaluate the diagnostic performance of Kaiser score (KS) adjusted with the apparent diffusion coefficient (ADC) (KS+) and machine learning (ML) modeling.

METHODS

A dataset of 402 malignant and 257 benign lesions was identified. Two radiologists assigned the KS. If a lesion with KS > 4 had ADC > 1.4 × 10^-3 mm²/s, the KS was reduced by 4 to become KS+. In order to consider the full spectrum of ADC as a continuous variable, the KS and ADC values were used to train diagnostic models using 5 ML algorithms. The performance was evaluated using the ROC analysis, compared by the DeLong test. The sensitivity, specificity, and accuracy achieved using the threshold of KS > 4, KS+ > 4, and ADC ≤ 1.4 × 10^-3 mm²/s were obtained and compared by the McNemar test.

RESULTS

The ROC curves of KS, KS+, and all ML models had comparable AUC in the range of 0.883-0.921, significantly higher than that of ADC (0.837, p < 0.0001). The KS had sensitivity = 97.3% and specificity = 59.1%; and the KS+ had sensitivity = 95.5% with significantly improved specificity to 68.5% (p < 0.0001). However, when setting at the same sensitivity of 97.3%, KS+ could not improve specificity. In ML analysis, the logistic regression model had the best performance. At sensitivity = 97.3% and specificity = 65.3%, i.e., compared to KS, 16 false-positives may be avoided without affecting true cancer diagnosis (p = 0.0015).

CONCLUSION

Using dichotomized ADC to modify KS to KS+ can improve specificity, but at the price of lowered sensitivity. Machine learning algorithms may be applied to consider the ADC as a continuous variable to build more accurate diagnostic models.

KEY POINTS

• When using ADC to modify the Kaiser score to KS+, the diagnostic specificity according to the results of two independent readers was improved by 9.4-9.7%, at the price of slightly degraded sensitivity by 1.5-1.8%, and overall had improved accuracy by 2.6-2.9%. • When the KS and the continuous ADC values were combined to train models by machine learning algorithms, the diagnostic specificity achieved by the logistic regression model could be significantly improved from 59.1 to 65.3% (p = 0.0015), while maintaining at the high sensitivity of KS = 97.3%, and thus, the results demonstrated the potential of ML modeling to further evaluate the contribution of ADC. • When setting the sensitivity at the same levels, the modified KS+ and the original KS have comparable specificity; therefore, KS+ with consideration of ADC may not offer much practical help, and the original KS without ADC remains as an excellent robust diagnostic method.

Magnetic resonance imaging correlation with molecular and epigenetic markers in assessment of breast cancer

Background

Breast cancer is a heterogeneous disease with a wide range of clinical behavior, histologic subtypes, therapeutic options, and outcomes. The different biology and histology of breast cancer display different tumor morphology at breast magnetic resonance imaging (MRI). However, few studies have examined the relationship between the MRI morphological, kinetic features and molecular and epigenetic markers in breast cancer assessment. The study aimed to evaluate the correlation between MRI morphological and kinetic features, molecular and an epigenetic marker (linc-ITGB1) in breast cancer cases. A total of 115 women (80 cases and 35 controls) with BIRAD 4 category breast lesions were included. The association between the MRI morphological & kinetic features, apparent diffusion coefficient (ADC) values, and molecular and the epigenetic marker (linc-ITGB1) was evaluated using Mann–Whitney and Chi-square tests.

Results

The shape (p = 0.009), size of the lesion (p = 0.003), and pattern of enhancement (p ≤ 0.001) were significantly correlated with the molecular markers. Luminal subtypes are more likely to be presented with irregular shaped and non-circumscribed margin masses (97% for luminal A and 94.1% for luminal B). Triple-negative cancers are frequently presented with regular masses, circumscribed margins, and peripheral rim enhancement (50% of TN tumors). HER2-positive cancers are more likely to be multifocal/multicentric and are more associated with non-mass pattern of enhancement as compared to HER2-negative cancers. Perilesional edema was also significantly correlated with HER2-positive lesions (p = 0.009). Although the epigenetic marker linc-ITGB1 was overexpressed by 4.85-folds in breast cancer cases compared to benign controls, we could not find any significant correlation between its expression level and the MRI features or molecular subtypes (p = 0.948).

Conclusions

MRI features can be a reliable predictor of breast cancer molecular subtypes. The epigenetic marker linc-ITGB1 has a potential role in breast cancer pathogenesis but with no significant correlation with either the MRI features or molecular subtypes of the lesions.

Can DWI provide additional value to Kaiser score in evaluation of breast lesions

OBJECTIVES

To explore added value of diffusion-weighted imaging (DWI) as an adjunct to Kaiser score (KS) for differentiation of benign from malignant lesions on breast magnetic resonance imaging (MRI).

METHODS

Two hundred forty-six patients with 273 lesions (155 malignancies) were included in this retrospective study from January 2015 to December 2019. All lesions were proved by pathology. Two radiologists blind to pathological results evaluated lesions according to KS. Lesions with score > 4 were considered malignant. Four thresholds of ADC values -1.3 × 10^-3mm²/s, 1.4 × 10^-3mm²/s, 1.53 × 10^-3mm²/s, and 1.6 × 10^-3mm²/s were used to distinguish benign from malignant lesions. For combined diagnosis, a lesion with KS > 4 and ADC values below the preset cutoffs was considered as malignant; otherwise, it was benign. Sensitivity, specificity, and area under the curve (AUC) were compared between KS, DWI, and combined diagnosis.

RESULTS

The AUC of KS was significantly higher than that of DWI alone (0.941 vs 0.901, p = 0.04). The sensitivity of KS (96.8%) and DWI (97.4 - 99.4%) was comparable (p > 0.05) while the specificity of KS (83.9%) was significantly higher than that of DWI (19.5-56.8%) (p < 0.05). Adding DWI as an adjunct to KS resulted in a 0-2.5% increase of specificity and a 0.1-1.3% decrease of sensitivity; however, the difference did not reach statistical significance (p > 0.05).

CONCLUSION

KS showed higher diagnostic performance than DWI alone for discrimination of breast benign and malignant lesions. DWI showed no additional value to KS for characterizing breast lesions.

KEY POINTS

• KS showed higher diagnostic performance than DWI alone for differentiation of benign from breast malignant lesions. • DWI alone showed a high sensitivity but a low specificity for characterizing breast lesions. • Diagnostic performance did not improve using DWI as an adjunct to KS.

Development and validation of a deep learning model for breast lesion segmentation and characterization in multiparametric MRI

Importance

The utilization of artificial intelligence for the differentiation of benign and malignant breast lesions in multiparametric MRI (mpMRI) assists radiologists to improve diagnostic performance.

Objectives

To develop an automated deep learning model for breast lesion segmentation and characterization and to evaluate the characterization performance of AI models and radiologists.

Materials and methods

For lesion segmentation, 2,823 patients were used for the training, validation, and testing of the VNet-based segmentation models, and the average Dice similarity coefficient (DSC) between the manual segmentation by radiologists and the mask generated by VNet was calculated. For lesion characterization, 3,303 female patients with 3,607 pathologically confirmed lesions (2,213 malignant and 1,394 benign lesions) were used for the three ResNet-based characterization models (two single-input and one multi-input models). Histopathology was used as the diagnostic criterion standard to assess the characterization performance of the AI models and the BI-RADS categorized by the radiologists, in terms of sensitivity, specificity, accuracy, and the area under the receiver operating characteristic curve (AUC). An additional 123 patients with 136 lesions (81 malignant and 55 benign lesions) from another institution were available for external testing.

Results

Of the 5,811 patients included in the study, the mean age was 46.14 (range 11–89) years. In the segmentation task, a DSC of 0.860 was obtained between the VNet-generated mask and manual segmentation by radiologists. In the characterization task, the AUCs of the multi-input and the other two single-input models were 0.927, 0.821, and 0.795, respectively. Compared to the single-input DWI or DCE model, the multi-input DCE and DWI model obtained a significant increase in sensitivity, specificity, and accuracy (0.831 vs. 0.772/0.776, 0.874 vs. 0.630/0.709, 0.846 vs. 0.721/0.752). Furthermore, the specificity of the multi-input model was higher than that of the radiologists, whether using BI-RADS category 3 or 4 as a cutoff point (0.874 vs. 0.404/0.841), and the accuracy was intermediate between the two assessment methods (0.846 vs. 0.773/0.882). For the external testing, the performance of the three models remained robust with AUCs of 0.812, 0.831, and 0.885, respectively.

Conclusions

Combining DCE with DWI was superior to applying a single sequence for breast lesion characterization. The deep learning computer-aided diagnosis (CADx) model we developed significantly improved specificity and achieved comparable accuracy to the radiologists with promise for clinical application to provide preliminary diagnoses.

Diagnostic performance of DCE-MRI, multiparametric MRI and multimodality imaging for discrimination of breast non-mass-like enhancement lesions

OBJECTIVE

The aim of this study was to investigate and compare the diagnostic performance of dynamic contrast-enhanced (DCE)-MRI, multiparametric MRI (mpMRI), and multimodality imaging (MMI) combining mpMRI and mammography (MG) for discriminating breast non-mass-like enhancement (NME) lesions.

METHODS

This retrospective study enrolled 193 patients with 199 lesions who underwent 3.0 T MRI and MG from January 2017 to December 2019. The features of DCE-MRI, turbo inversion recovery magnitude (TIRM), and diffusion-weighted imaging (DWI) were assessed by two breast radiologists. Then, all lesions were divided into microcalcification and non-microcalcification groups to assess the features of MG. Comparisons were performed between groups using univariate analyses. Then, multivariate analyses were performed to construct diagnostic models for distinguishing NME lesions. Diagnostic performance was evaluated by using the area under the curve (AUC) and the differences between AUCs were evaluated by using the DeLong test.

RESULTS

Overall (n = 199), mpMRI outperformed DCE-MRI alone (AUCmpMRI = 0.924 vs. AUCDCE-MRI = 0.884; p = 0.007). Furthermore, MMI outperformed both mpMRI and MG (the microcalcification group [n = 140]: AUCMMI = 0.997 vs. AUCmpMRI = 0.978, p = 0.018 and AUCMMI = 0.997 vs. AUCMG = 0.912, p < 0.001; the non-microcalcification group [n = 59]: AUCMMI = 0.857 vs. AUCmpMRI = 0.768, p = 0.044 and AUCMMI = 0.857 vs. AUCMG = 0.759, p = 0.039).

CONCLUSION & ADVANCES IN KNOWLEDGE

DCE-MRI combined with DWI and TIRM information could improve the diagnostic performance for discriminating NME lesions compared with DCE-MRI alone. Furthermore, MMI combining mpMRI and MG showed better discrimination than both mpMRI and MG.

A simplified scoring protocol to improve diagnostic accuracy with the breast imaging reporting and data system in breast magnetic resonance imaging

BACKGROUND

The breast imaging reporting and data system (BI-RADS) lexicon provides a standardized terminology for describing leision characteristics but does not provide defined rules for converting specific imaging features into diagnostic categories. The inter-reader agreement of the BI-RADS is moderate. In this study, we explored the use of a simplified protocol and scoring system for BI-RADS categorization which integrates the morphologic features (MF), kinetic time-intensity curve (TIC), and apparent diffusion coefficient (ADC) values with equal weights, with a view to providing a convenient and practical method for breast magnetic resonance imaging (MRI) and improving the inter-reader agreement and diagnostic performance of BI-RADS.

METHODS

This cross-sectional, retrospective, single-center study included 879 patients with 898 histopathologically verified lesions who underwent an MRI scan on a 3.0 Tesla GE Discovery 750 MRI scanner between January 1, 2017, and June 30, 2020. The BI-RADS categorization of the studied lesions was assessed according to the sum of the assigned scores (the presence of malignant MF, lower ADC, and suspicious TIC each warranted a score of +1). Total scores of +2 and +3 were classified as category 5, scores of +1 were classified as category 4, and scores of +0 but with other lesions of interest were classified as category 3. The receiver operating characteristic (ROC) curves were plotted, and the sensitivity, specificity, and accuracy of this categorization were investigated to assess its efficacy and its consistency with pathology.

RESULTS

There were 472 malignant, 104 risk, and 322 benign lesions. Our simplified scoring protocol had high diagnostic accuracy, with an area under curve (AUC) value of 0.896. In terms of the borderline effect of pathological risk and category 4 lesions, our results showed that when risk lesions were classified together with malignant ones, the AUC value improved (0.876 vs. 0.844 and 0.909 vs. 0.900). When category 4 and 5 lesions were classified as malignant, the specificity, accuracy, and AUC value decreased (82.3% vs. 93.2%, 89.3% vs. 90.2%, and 0.876 vs. 0.909, respectively). Therefore, to improve the diagnostic accuracy of the protocol for BI-RADS categorization, only category 5 lesions should be considered to be malignant.

CONCLUSIONS

Our simplified scoring protocol that integrates MF, TIC, and ADC values with equal weights for BI-RADS categorization could improve both the diagnostic performance of the protocol for BI-RADS categorization in clinical practice and the understanding of the benign-risk-malignant breast diseases.

Breast MRI during pregnancy and lactation: clinical challenges and technical advances

The breast experiences substantial changes in morphology and function during pregnancy and lactation which affects its imaging properties and may reduce the visibility of a concurrent pathological process. The high incidence of benign gestational-related entities may further add complexity to the clinical and radiological evaluation of the breast during the period. Consequently, pregnancy-associated breast cancer (PABC) is often a delayed diagnosis and carries a poor prognosis. This state-of-the-art pictorial review illustrates how despite currently being underutilized, technical advances and new clinical evidence support the use of unenhanced breast MRI during pregnancy and both unenhanced and dynamic-contrast enhanced (DCE) during lactation, to serve as effective supplementary modalities in the diagnostic work-up of PABC.

Breast Lesion Classification with Multiparametric Breast MRI Using Radiomics and Machine Learning: A Comparison with Radiologists' Performance

Simple Summary

Currently, breast contrast-enhanced MRI is the most sensitive imaging technique for breast cancer detection; however, its specificity is low given the common characteristics shared by benign breast lesions and some cancers. This leads to a high number of false-positive cases and, therefore, unnecessary biopsies. Multiparametric MRI including diffusion-weighted imaging assists in this task by increasing the specificity for breast lesion discrimination. Nevertheless, interpretation of breast MRI is still highly dependent on the reader’s level of experience. Our work combines radiomic features extracted from multiparametric MRI to generate predictive models for breast cancer differentiation. Additionally, decision support models were compared with the performance of two breast dedicated radiologists for lesion differentiation. Our work proves the potential of multiparametric radiomics coupled with machine learning to be implemented in clinical practice for lesion differentiation on breast MRI. AI algorithms show value to assist less experienced readers, improving the accuracy for breast lesion discrimination.

Abstract

This multicenter retrospective study compared the performance of radiomics analysis coupled with machine learning (ML) with that of radiologists for the classification of breast tumors. A total of 93 consecutive women (mean age: 49 ± 12 years) with 104 histopathologically verified enhancing lesions (mean size: 22.8 ± 15.1 mm), classified as suspicious on multiparametric breast MRIs were included. Two experienced breast radiologists assessed all of the lesions, assigning a Breast Imaging Reporting and Database System (BI-RADS) suspicion category, providing a diffusion-weighted imaging (DWI) score based on lesion signal intensity, and determining the apparent diffusion coefficient (ADC). Ten predictive models for breast lesion discrimination were generated using radiomic features extracted from the multiparametric MRI. The area under the receiver operating curve (AUC) and the accuracy were compared using McNemar’s test. Multiparametric radiomics with DWI score and BI-RADS (accuracy = 88.5%; AUC = 0.93) and multiparametric radiomics with ADC values and BI-RADS (accuracy= 88.5%; AUC = 0.96) models showed significant improvements in diagnostic accuracy compared to the multiparametric radiomics (DWI + DCE data) model (p = 0.01 and p = 0.02, respectively), but performed similarly compared to the multiparametric assessment by radiologists (accuracy = 85.6%; AUC = 0.03; p = 0.39). In conclusion, radiomics analysis coupled with the ML of multiparametric MRI could assist in breast lesion discrimination, especially for less experienced readers of breast MRIs.

Multiparametric Magnetic Resonance Imaging in Evaluation of Benign and Malignant Breast Masses with Pathological Correlation

Background

Dynamic contrast-enhanced (DCE) MRI sequences plays a vital role in diagnosing breast masses with high sensitivity and specificity as compared to other diagnostic modalities. The addition of diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) values significantly improves diagnostic accuracy. This study aimed to study the breast masses on DCE-MRI, restricted diffusion on DWI, ADC values, and choline peak on spectroscopy in breast cancer diagnosis.

Material and methods

This study was a prospective observational study which involved subjects with breast lumps. Baseline data was collected from the patients along with pertinent clinical history and relevant laboratory investigations. MR mammography (MRM) was performed on a 1.5 Tesla MR Scanner (MAGNETOM® Avanto, Siemens AG, Munich Germany) using a dedicated double breast coil.

Results

Forty-one subjects were included with a total of 54 breast masses in them. The mean age of the study population was 47.1±14.7 years. From the MRI final diagnosis, the majority (53.70%) were diagnosed as malignant lesions and 46.30% as benign. Out of 20 lesions diagnosed as benign on histopathology, only five percent had ADC value <1.3 ×10⁻³mm²/s, and the majority (95%) had ADC value >1.3 ×10⁻³mm²/s. All 20 lesions were circumscribed, ovoid, or round in shape showing no restricted diffusion on DWI, with corresponding ADC value of >1.3×10⁻³mm²/s, homogeneous post-contrast enhancement, or with dark internal septations, type I kinetic enhancement curve, and they showed no choline peak on spectroscopy. Out of 34 malignant lesions diagnosed on histopathology, the majority (85.29%) displayed restricted diffusion on DWI and had an ADC value of <1.3×10⁻³mm²/s, most of them had spiculated margins, type II/ III kinetic curve with choline peak on spectroscopy.

Conclusion

Multiparametric MR mammography, which included DCE-MRM, DWI, ADC values, and spectroscopy, correlated well with the histopathological diagnosis of benign and malignant breast masses.

Downgrade BI-RADS 4A Patients Using Nomogram Based on Breast Magnetic Resonance Imaging, Ultrasound, and Mammography

Objectives

To downgrade BI-RADS 4A patients by constructing a nomogram using R software.

Materials and Methods

A total of 1,717 patients were retrospectively analyzed who underwent preoperative ultrasound, mammography, and magnetic resonance examinations in our hospital from August 2019 to September 2020, and a total of 458 patients of category BI-RADS 4A (mean age, 47 years; range 18–84 years; all women) were included. Multivariable logistic regression was used to screen out the independent influencing parameters that affect the benign and malignant tumors, and the nomogram was constructed by R language to downgrade BI-RADS 4A patients to eligible category.

Results

Of 458 BI-RADS 4A patients, 273 (59.6%) were degraded to category 3. The malignancy rate of these 273 lesions is 1.5% (4/273) (<2%), and the sensitivity reduced to 99.6%, the specificity increased from 4.41% to 45.3%, and the accuracy increased from 63.4% to 78.8%.

Conclusion

By constructing a nomogram, some patients can be downgraded to avoid unnecessary biopsy.

Breast Cancer Classification on Multiparametric MRI - Increased Performance of Boosting Ensemble Methods

Introduction: This study aims to assess the utility of Boosting ensemble classification methods for increasing the diagnostic performance of multiparametric Magnetic Resonance Imaging (mpMRI) radiomic models, in differentiating benign and malignant breast lesions. Methods: The dataset includes mpMR images of 140 female patients with mass-like breast lesions (70 benign and 70 malignant), consisting of Dynamic Contrast Enhanced (DCE) and T2-weighted sequences, and the Apparent Diffusion Coefficient (ADC) calculated from the Diffusion Weighted Imaging (DWI) sequence. Tumor masks were manually defined in all consecutive slices of the respective MRI volumes and 3D radiomic features were extracted with the Pyradiomics package. Feature dimensionality reduction was based on statistical tests and the Boruta wrapper. Hierarchical Clustering on Spearman's rank correlation coefficients between features and Random Forest classification for obtaining feature importance, were implemented for selecting the final feature subset. Adaptive Boosting (AdaBoost), Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM) classifiers, were trained and tested with bootstrap validation in differentiating breast lesions. A Support Vector Machine (SVM) classifier was also exploited for comparison. The Receiver Operator Characteristic (ROC) curves and DeLong's test were utilized to evaluate the classification performances. Results: The final feature subset consisted of 5 features derived from the lesion shape and the first order histogram of DCE and ADC images volumes. XGboost and LGBM achieved statistically significantly higher average classification performances [AUC = 0.95 and 0.94 respectively], followed by Adaboost [AUC = 0.90], GB [AUC = 0.89] and SVM [AUC = 0.88]. Conclusion: Overall, the integration of Ensemble Learning methods within mpMRI radiomic analysis can improve the performance of computer-assisted diagnosis of breast cancer lesions.

A Comparative Assessment of MR BI-RADS 4 Breast Lesions With Kaiser Score and Apparent Diffusion Coefficient Value

Objectives

To investigate the diagnostic performance of the Kaiser score and apparent diffusion coefficient (ADC) to differentiate Breast Imaging Reporting and Data System (BI-RADS) Category 4 lesions at dynamic contrast-enhanced (DCE) MRI.

Methods

This was a single-institution retrospective study of patients who underwent breast MRI from March 2020 to June 2021. All image data were acquired with a 3-T MRI system. Kaiser score of each lesion was assigned by an experienced breast radiologist. Kaiser score+ was determined by combining ADC and Kaiser score. Receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic performance of Kaiser score+, Kaiser score, and ADC. The area under the curve (AUC) values were calculated and compared by using the Delong test. The differences in sensitivity and specificity between different indicators were determined by the McNemar test.

Results

The study involved 243 women (mean age, 43.1 years; age range, 18-67 years) with 268 MR BI-RADS 4 lesions. Overall diagnostic performance for Kaiser score (AUC, 0.902) was significantly higher than for ADC (AUC, 0.81; p = 0.004). There were no significant differences in AUCs between Kaiser score and Kaiser score+ (p = 0.134). The Kaiser score was superior to ADC in avoiding unnecessary biopsies (p < 0.001). Compared with the Kaiser score alone, the specificity of Kaiser score+ increased by 7.82%, however, at the price of a lower sensitivity.

Conclusion

For MR BI-RADS category 4 breast lesions, the Kaiser score was superior to ADC mapping regarding the potential to avoid unnecessary biopsies. However, the combination of both indicators did not significantly contribute to breast cancer diagnosis of this subgroup.

Application of MRI Image Based on Computer Semiautomatic Segmentation Algorithm in the Classification Prediction of Breast Cancer Histology

Objective

The study aimed to investigate the predictive classification accuracy of computer semiautomatic segmentation algorithm for the histological grade of breast tumors through the magnetic resonance imaging (MRI) examination.

Methods

Five dynamic contrast-enhanced (DCE) MRI regions of interest (ROIs) were captured using computer semiautomatic segmentation method, referring to the entire tumor area, tumor border area, proximal gland area, middle gland area, and distal gland area. According to the mutual information maximum protocol, the corresponding five ROIs were extracted from diffusion weighted imaging (DWI) combined with DCE-MRI images. To use the features in the nonoverlapping area of DWI image and DCE-MRI image as elements, a single-variable logistic regression model was established corresponding to element characteristics. After multiple training, the model was evaluated using the receiver operating characteristic (ROC) curve and area under curve (AUC).

Results

This DCE-MRI combined with DWI was superior to DCE-MRI and DW in the prediction of tumor area features. To use DCE-MRI or DWI alone was less effective than DCE-MRI combined with DWI. The DWI combined DCE-MRI demonstrated good regional segmentation effects in the tumour area, with luminal A value being 0.767 and the area under curve (AUC) value being 0.758. After optimization, the AUC value of the tumor area was 0.929, indicating that classification effects can be enhanced by combining the two imaging methods, which complemented each other.

Conclusions

The DWI combined DCE-MRI imaging has improved the early diagnosis effects of breast cancer by predicting the occurrence of breast cancer through the labeling of biomarkers.

Multiparameter MRI Model With DCE-MRI, DWI, and Synthetic MRI Improves the Diagnostic Performance of BI-RADS 4 Lesions

Objectives

To evaluate the value of synthetic magnetic resonance imaging (syMRI), diffusion-weighted imaging (DWI), DCE-MRI, and clinical features in breast imaging–reporting and data system (BI-RADS) 4 lesions, and develop an efficient method to help patients avoid unnecessary biopsy.

Methods

A total of 75 patients with breast diseases classified as BI-RADS 4 (45 with malignant lesions and 30 with benign lesions) were prospectively enrolled in this study. T1-weighted imaging (T1WI), T2WI, DWI, and syMRI were performed at 3.0 T. Relaxation time (T1 and T2), apparent diffusion coefficient (ADC), conventional MRI features, and clinical features were assessed. “T” represents the relaxation time value of the region of interest pre-contrast scanning, and “T+” represents the value post-contrast scanning. The rate of change in the T value between pre- and post-contrast scanning was represented by ΔT%.

Results

ΔT1%, T2, ADC, age, body mass index (BMI), menopause, irregular margins, and heterogeneous internal enhancement pattern were significantly associated with a breast cancer diagnosis in the multivariable logistic regression analysis. Based on the above parameters, four models were established: model 1 (BI-RADS model, including all conventional MRI features recommended by BI-RADS lexicon), model 2 (relaxation time model, including ΔT1% and T2), model 3 [multi-parameter (mp)MRI model, including ΔT1%, T2, ADC, margin, and internal enhancement pattern], and model 4 (combined image and clinical model, including ΔT1%, T2, ADC, margin, internal enhancement pattern, age, BMI, and menopausal state). Among these, model 4 has the best diagnostic performance, followed by models 3, 2, and 1.

Conclusions

The mpMRI model with DCE-MRI, DWI, and syMRI is a robust tool for evaluating the malignancies in BI-RADS 4 lesions. The clinical features could further improve the diagnostic performance of the model.

Transfer Learning Strategy Based on Unsupervised Learning and Ensemble Learning for Breast Cancer Molecular Subtype Prediction Using Dynamic Contrast-Enhanced MRI

BACKGROUND

Imaging-driven deep learning strategies focus on training from scratch and transfer learning. However, the performance of training from scratch is often impeded by the lack of large-scale labeled training data. Additionally, owing to the differences between source and target domains, analyzing medical image tasks satisfactorily via transfer learning based on ImageNet is difficult.

PURPOSE

To investigate two transfer learning algorithms for breast cancer molecular subtype prediction (luminal and non-luminal) based on unsupervised pre-training and ensemble learning: M_EL and B_EL, using malignant and benign datasets as the source domain, respectively.

STUDY TYPE

Retrospective.

POPULATION

Eight hundred and thirty-three female patients with histologically confirmed breast lesions (567 benign and 266 malignant cases) were selected. In the 5-fold cross-validation, the malignant cohort was randomly divided into 5 subsets to form a training set (80%) and a validation set (20%).

FIELD STRENGTH/SEQUENCE

3.0 T, dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) using T1-weighted high-resolution isotropic volume examination.

ASSESSMENT

First, three datasets acquired at different times post-contrast were preprocessed as unlabeled source domains. Second, three baseline networks corresponding to the different MRI post-contrast phases were built, optimized by a combination of mutual information maximization between high- and low-level representations and prior distribution constraints. Next, the pre-trained networks were fine-tuned on the labeled target domain. Finally, prediction results were integrated using weighted voting-based ensemble learning.

STATISTICAL TESTS

Mean accuracy, precision, specificity, and area under receiver operating characteristic curve (AUC) were obtained with 5-fold cross-validation. P < 0.05 was considered to be statistically significant.

RESULTS

Compared with a convolutional long short-term memory network, pre-trained VGG-16, VGG-19, and DenseNet-121 from ImageNet, M_EL and B_EL exhibited significantly more optimized prediction performance (specificity: 90.5% and 89.9%; accuracy: 82.6% and 81.1%; precision: 91.2% and 90.9%; AUC: 0.836 and 0.823, respectively).

DATA CONCLUSION

Transfer learning based on unsupervised pre-training may facilitate automatic prediction of breast cancer molecular subtypes.

LEVEL OF EVIDENCE

3 TECHNICAL EFFICACY: Stage 2.

Design of a high-performance non-linear gradient coil for diffusion weighted MRI of the breast

Diffusion-weighted imaging (DWI) in the female breast is a magnetic resonance imaging (MRI) technique that complements clinical routine protocols, and that might provide an independent diagnostic value for specific clinical tasks in breast imaging. To further improve specificity of DWI in the breast, stronger and faster diffusion weighting is advantageous. Here, a dedicated gradient coil is designed, targeted at diffusion weighting in the female breast, with the peak gradient magnitude exceeding that of the current clinical MR scanners by an order of a magnitude. Design of application-tailored gradient coils in MRI has recently attracted increased attention. With the target application in mind, the gradient coil is designed on an irregularly shaped semi-open current-carrying surface. Due to the coil former closely fitting the non-spherical target region, non-linear encoding fields become particularly advantageous for achieving locally exceptionally high gradient strengths. As breast tissue has a predominantly isotropic cellular microstructure, the direction of the diffusion-weighting gradient may be allowed to vary within the target volume. However, due to the quadratic dependency of the b-factor on the gradient strength, variation of the gradient magnitude should be carefully controlled. To achieve the above design goals the corresponding multi-objective optimization problem is reformulated as a constrained optimization, allowing for flexible and precise control of the coil properties. A novel constraint is proposed, limiting gradient magnitude variation within every slice while allowing for variations in both the direction of the gradient within the slice and the magnitude across the slices. These innovations enable the design of a unilateral coil for diffusion weighting in the female breast with local gradient strengths exceeding 1 T/m with highly homogeneous diffusion weighting for imaging in the coronal slice orientation.

What's Hot in Breast MRI

Several articles in the literature have demonstrated a promising role for breast MRI techniques that are more economic in total exam time than others when used as supplement to mammography for detection and diagnosis of breast cancer. There are many technical factors that must be considered in the shortened breast MRI protocols to cut down time of standard ones, including using optimal fat suppression, gadolinium-chelates intravascular contrast administrations for dynamic imaging with post processing subtractions and maximum intensity projections (MIP) high spatial and temporal resolution among others. Multiparametric breast MRI that includes both gadolinium-dependent, i.e., dynamic contrast-enhanced (DCE-MRI) and gadolinium-free techniques, i.e., diffusion-weighted/diffusion-tensor magnetic resonance imaging (DWI/DTI) are shown by several investigators that can provide extremely high sensitivity and specificity for detection of breast cancer. This article provides an overview of the proven indications for breast MRI including breast cancer screening for higher than average risk, determining chemotherapy induced tumor response, detecting residual tumor after incomplete surgical excision, detecting occult cancer in patients presenting with axillary node metastasis, detecting residual tumor after incomplete breast cancer surgical excision, detecting cancer when results of conventional imaging are equivocal, as well patients suspicious of having breast implant rupture. Despite having the highest sensitivity for breast cancer detection, there are pitfalls, however, secondary to false positive and false negative contrast enhancement and contrast-free MRI techniques. Awareness of the strengths and limitations of different approaches to obtain state of the art MR images of the breast will facilitate the work-up of patients with suspicious breast lesions.

Discrimination of Malignant and Benign Breast Lesions Using Quantitative Multiparametric MRI: A Preliminary Study

We aimed to compare diagnostic performance in discriminating malignant and benign breast lesions between two intravoxel incoherent motion (IVIM) analysis methods for diffusion-weighted magnetic resonance imaging (DW-MRI) data and between DW- and dynamic contrast-enhanced (DCE)-MRI, and to determine if combining DW- and DCE-MRI further improves diagnostic accuracy. DW-MRI with 12 b-values and DCE-MRI were performed on 26 patients with 28 suspicious breast lesions before biopsies. The traditional biexponential fitting and a 3-b-value method were used for independent IVIM analysis of the DW-MRI data. Simulations were performed to evaluate errors in IVIM parameter estimations by the two methods across a range of signal-to-noise ratio (SNR). Pharmacokinetic modeling of DCE-MRI data was performed. Conventional radiological MRI reading yielded 86% sensitivity and 21% specificity in breast cancer diagnosis. At the same sensitivity, specificity of individual DCE- and DW-MRI markers improved to 36%–57% and that of combined DCE- or combined DW-MRI markers to 57%–71%, with DCE-MRI markers showing better diagnostic performance. The combination of DCE- and DW-MRI markers further improved specificity to 86%–93% and the improvements in diagnostic accuracy were statistically significant (P < .05) when compared with standard clinical MRI reading and most individual markers. At low breast DW-MRI SNR values (<50), like those typically seen in clinical studies, the 3-b-value approach for IVIM analysis generates markers with smaller errors and with comparable or better diagnostic performances compared with biexponential fitting. This suggests that the 3-b-value method could be an optimal IVIM-MRI method to be combined with DCE-MRI for improved diagnostic accuracy.

Radiomics and Machine Learning with Multiparametric Breast MRI for Improved Diagnostic Accuracy in Breast Cancer Diagnosis

The purpose of this multicenter retrospective study was to evaluate radiomics analysis coupled with machine learning (ML) of dynamic contrast-enhanced (DCE) and diffusion-weighted imaging (DWI) radiomics models separately and combined as multiparametric MRI for improved breast cancer detection. Consecutive patients (Memorial Sloan Kettering Cancer Center, January 2018-March 2020; Medical University Vienna, from January 2011-August 2014) with a suspicious enhancing breast tumor on breast MRI categorized as BI-RADS 4 and who subsequently underwent image-guided biopsy were included. In 93 patients (mean age: 49 years ± 12 years; 100% women), there were 104 lesions (mean size: 22.8 mm; range: 7-99 mm), 46 malignant and 58 benign. Radiomics features were calculated. Subsequently, the five most significant features were fitted into multivariable modeling to produce a robust ML model for discriminating between benign and malignant lesions. A medium Gaussian support vector machine (SVM) model with five-fold cross validation was developed for each modality. A model based on DWI-extracted features achieved an AUC of 0.79 (95% CI: 0.70-0.88), whereas a model based on DCE-extracted features yielded an AUC of 0.83 (95% CI: 0.75-0.91). A multiparametric radiomics model combining DCE- and DWI-extracted features showed the best AUC (0.85; 95% CI: 0.77-0.92) and diagnostic accuracy (81.7%; 95% CI: 73.0-88.6). In conclusion, radiomics analysis coupled with ML of multiparametric MRI allows an improved evaluation of suspicious enhancing breast tumors recommended for biopsy on clinical breast MRI, facilitating accurate breast cancer diagnosis while reducing unnecessary benign breast biopsies.

3T MRI-Radiomic Approach to Predict for Lymph Node Status in Breast Cancer Patients

Simple Summary

Breast cancer is the most common cancer in women worldwide. The axillary lymph node status is one of the main prognostic factors. Currently, the methods to define the lymph node status are invasive and not without sequelae (from biopsy to lymphadenectomy). Radiomics is a new tool, and highly varied, but with high potential that has already shown excellent results in numerous fields of application. In our study, we have developed a classifier validated on a relatively large number of patients, which is able to predict lymph node status using a combination of patients clinical features, primary breast cancer histological features and radiomics features based on 3 Tesla post contrast—MR images. This approach can accurately select breast cancer patients who may avoid unnecessary biopsy and lymphadenectomy in a non-invasive way.

Abstract

Background: axillary lymph node (LN) status is one of the main breast cancer prognostic factors and it is currently defined by invasive procedures. The aim of this study is to predict LN metastasis combining MRI radiomics features with primary breast tumor histological features and patients’ clinical data. Methods: 99 lesions on pre-treatment contrasted 3T-MRI (DCE). All patients had a histologically proven invasive breast cancer and defined LN status. Patients’ clinical data and tumor histological analysis were previously collected. For each tumor lesion, a semi-automatic segmentation was performed, using the second phase of DCE-MRI. Each segmentation was optimized using a convex-hull algorithm. In addition to the 14 semantics features and a feature ROI volume/convex-hull volume, 242 other quantitative features were extracted. A wrapper selection method selected the 15 most prognostic features (14 quantitative, 1 semantic), used to train the final learning model. The classifier used was the Random Forest. Results: the AUC-classifier was 0.856 (label = positive or negative). The contribution of each feature group was lower performance than the full signature. Conclusions: the combination of patient clinical, histological and radiomics features of primary breast cancer can accurately predict LN status in a non-invasive way.

Noncontrast-Enhanced MR-Based Conductivity Imaging for Breast Cancer Detection and Lesion Differentiation

BACKGROUND

There is increasing interest in noncontrast-enhanced MRI due to safety concerns for gadolinium contrast agents.

PURPOSE

To investigate the clinical feasibility of MR-based conductivity imaging for breast cancer detection and lesion differentiation.

STUDY TYPE

Prospective.

SUBJECTS

One hundred and ten women, with 112 known cancers and 17 benign lesions (biopsy-proven), scheduled for preoperative MRI.

FIELD STRENGTH/SEQUENCE

Non-fat-suppressed T2-weighted turbo spin-echo sequence (T2WI), dynamic contrast-enhanced MRI and diffusion-weighted imaging (DWI) at 3T.

ASSESSMENT

Cancer detectability on each imaging modality was qualitatively evaluated on a per-breast basis: the conductivity maps derived from T2WI were independently reviewed by three radiologists (R1-R3). T2WI, DWI, and pre-operative digital mammography were independently reviewed by three other radiologists (R4-R6). Conductivity and apparent diffusion coefficient (ADC) measurements (mean, minimum, and maximum) were performed for 112 cancers and 17 benign lesions independently by two radiologists (R1 and R2). Tumor size was measured from surgical specimens.

STATISTICAL TESTS

Cancer detection rates were compared using generalized estimating equations. Multivariable logistic regression analysis was performed to identify factors associated with cancer detectability. Discriminating ability of conductivity and ADC was evaluated by using the areas under the receiver operating characteristic curve (AUC).

RESULTS

Conductivity imaging showed lower cancer detection rates (20%-32%) compared to T2WI (62%-71%), DWI (85%-90%), and mammography (79%-88%) (all P < 0.05). Fatty breast on MRI (odds ratio = 11.8, P < 0.05) and invasive tumor size (odds ratio = 1.7, P < 0.05) were associated with cancer detectability of conductivity imaging. The maximum conductivity showed comparable ability to the mean ADC in discriminating between cancers and benign lesions (AUC = 0.67 [95% CI: 0.59, 0.75] vs. 0.84 [0.76, 0.90], P = 0.06 (R1); 0.65 [0.56, 0.73] vs. 0.82 [0.74, 0.88], P = 0.07 (R2)).

DATA CONCLUSION

Although conductivity imaging showed suboptimal performance in breast cancer detection, the quantitative measurement of conductivity showed the potential for lesion differentiation.

EVIDENCE LEVEL

1 TECHNICAL EFFICACY: Stage 2.

Machine Learning Based on Multi-Parametric MRI to Predict Risk of Breast Cancer

Purpose

Machine learning (ML) can extract high-throughput features of images to predict disease. This study aimed to develop nomogram of multi-parametric MRI (mpMRI) ML model to predict the risk of breast cancer.

Methods

The mpMRI included non-enhanced and enhanced T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), apparent diffusion coefficient (ADC), K ^trans, K _ep, V _e, and V _p. Regions of interest were annotated in an enhanced T1WI map and mapped to other maps in every slice. 1,132 features and top-10 principal components were extracted from every parameter map. Single-parametric and multi-parametric ML models were constructed via 10 rounds of five-fold cross-validation. The model with the highest area under the curve (AUC) was considered as the optimal model and validated by calibration curve and decision curve. Nomogram was built with the optimal ML model and patients' characteristics.

Results

This study involved 144 malignant lesions and 66 benign lesions. The average age of patients with benign and malignant lesions was 42.5 years old and 50.8 years old, respectively, which were statistically different. The sixth and fourth principal components of K ^trans had more importance than others. The AUCs of K ^trans, K _ep, V _e and V _p, non-enhanced T1WI, enhanced T1WI, T2WI, and ADC models were 0.86, 0.81, 0.81, 0.83, 0.79, 0.81, 0.84, and 0.83 respectively. The model with an AUC of 0.90 was considered as the optimal model which was validated by calibration curve and decision curve. Nomogram for the prediction of breast cancer was built with the optimal ML models and patient age.

Conclusion

Nomogram could improve the ability of breast cancer prediction preoperatively.

Investigation of Synthetic Relaxometry and Diffusion Measures in the Differentiation of Benign and Malignant Breast Lesions as Compared to BI-RADS

BACKGROUND

Breast cancer is the most common malignant tumor in women and a quantitative contrast-free method is highly desirable for its diagnosis.

PURPOSE

To investigate the performance of quantitative MRI in differentiating malignant from benign breast lesions and to compare with the Breast Imaging Reporting and Data System (BI-RADS).

STUDY TYPE

Retrospective.

SUBJECTS

Eighty patients (56 with malignant lesions and 24 with benign lesions).

FIELD STRENGTH/SEQUENCE

Diffusion-weighted imaging (DWI) with a single-shot echo planar sequence and synthetic MRI with magnetic resonance image compilation (MAGiC) were performed at 3T.

ASSESSMENT

T₁ relaxation time (T₁ ), T₂ relaxation time (T₂ ), and proton density (PD) from synthetic MRI and apparent diffusion coefficient (ADC) from DWI were analyzed by two radiologists (Reader A, Reader B). Univariable and multivariable models were developed to optimize differentiation between malignant and benign lesions and their performances compared to BI-RADS.

STATISTICAL TESTS

The diagnostic performance was evaluated using multivariate logistic regression analysis and area under the receiver operating characteristic (ROC) curves (AUC).

RESULTS

T₂ , PD, and ADC values for malignant lesions were significantly lower than those in benign breast lesions for both radiologists (all P < 0.05). The combined T₂ , PD, and ADC model had the best performance for differentiating malignant and benign lesions with AUC, sensitivity, specificity, positive predictive value, and negative predictive values of 0.904, 94.6%, 87.5%, 94.6%, and 87.5%, respectively. The corresponding results for BI-RADS were no AUC, 94.6%, 75.0%, 89.8%, and 85.7%, respectively.

DATA CONCLUSION

The approach that combined synthetic MRI and DWI outperformed BI-RADS in the differential diagnosis of malignant and benign breast lesions and was achieved without contrast agents. This approach may serve as an alternative and effective strategy for the improvement of breast lesion differentiation.

LEVEL OF EVIDENCE

3.

TECHNICAL EFFICACY STAGE

3.

Influence of residual fat signal on diffusion kurtosis MRI of suspicious mammography findings

Recent studies showed the potential of diffusion kurtosis imaging (DKI) as a tool for improved classification of suspicious breast lesions. However, in diffusion-weighted imaging of the female breast, sufficient fat suppression is one of the main factors determining the success. In this study, the data of 198 patients examined in two study centres was analysed using standard diffusion and kurtosis evaluation methods and three DKI fitting approaches accounting phenomenologically for fat-related signal contamination of the lesions. Receiver operating characteristic curve analysis showed the highest area under the curve (AUC) for the method including fat correction terms (AUC = 0.85, p < 0.015) in comparison to the values obtained with the standard diffusion (AUC = 0.77) and kurtosis approach (AUC = 0.79). Comparing the two study centres, the AUC value improved from 0.77 to 0.86 (p = 0.036) using a fat correction term for the first centre, while no significant difference with no adverse effects was observed for the second centre (AUC 0.89 vs. 0.90, p = 0.95). Contamination of the signal in breast lesions with unsuppressed fat causing a reduction of diagnostic performance of diffusion kurtosis imaging may potentially be counteracted by proposed adapted evaluation methods.

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.

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.

Electronic supplementary material

The online version of this article (10.1007/s00330-020-06991-7) contains supplementary material, which is available to authorized users.

Free-breathing dynamic contrast-enhanced magnetic resonance of interstitial lung fibrosis

PURPOSE

Computed tomography (CT) imaging is the standard to assess interstitial lung disease. Magnetic resonance (MR) is potentially advantageous due to superior tissue characterization and better assessment of blood flow dynamics. This study aimed to evaluate idiopathic pulmonary fibrosis (IPF) using prototype 4D Stack of Stars GRE (StarVIBE) MR and compare it to CT.

METHOD

This IRB-approved prospective study included 13 patients [5F:8M; average age 66 ± 8.1 years] with pulmonary fibrosis, and 12 healthy controls [3F:9M; average age 55 ± 3.6 years]. MR of the chest included noncontrast steady-state free precession imaging (SSFP) and free-breathing 4D StarVIBE sequence with intravenous contrast administration up to 160 s. The images were assessed for quality and artifacts. The image resolution was evaluated based on the visibility of the smallest bronchi, vessels, lymph nodes, and pleural fissures. Independent assessment of reticulation, ground-glass opacity, and traction bronchiectasis was performed and compared to CT.

RESULTS

The StarVIBE images had fewer artifacts and higher spatial resolution. The findings associated with IPF were significantly better seen with StarVIBE, with superior CT correlation.

CONCLUSION

Contrast-enhanced free-breathing StarVIBE MR can generate high quality images with good correlation to CT in patients with IPF, and with high spatial and temporal resolution to generate rapid sequential dynamic images.

Artificial Intelligence-Based Classification of Breast Lesions Imaged With a Multiparametric Breast MRI Protocol With Ultrafast DCE-MRI, T2, and DWI

OBJECTIVES

We investigated artificial intelligence (AI)-based classification of benign and malignant breast lesions imaged with a multiparametric breast magnetic resonance imaging (MRI) protocol with ultrafast dynamic contrast-enhanced MRI, T2-weighted, and diffusion-weighted imaging with apparent diffusion coefficient mapping.

MATERIALS AND METHODS

We analyzed 576 lesions imaged with MRI, including a consecutive set of biopsied malignant (368) and benign (149) lesions, and an additional set of 59 benign lesions proven by follow-up. We used deep learning methods to interpret ultrafast dynamic contrast-enhanced MRI and T2-weighted information. A random forests classifier combined the output with patient information (PI; age and BRCA status) and apparent diffusion coefficient values obtained from diffusion-weighted imaging to perform the final lesion classification. We used receiver operating characteristic (ROC) analysis to evaluate our results. Sensitivity and specificity were compared with the results of the prospective clinical evaluation by radiologists.

RESULTS

The area under the ROC curve was 0.811 when only ultrafast dynamics was used. The final AI system that combined all imaging information with PI resulted in an area under the ROC curve of 0.852, significantly higher than the ultrafast dynamics alone (P = 0.002). When operating at the same sensitivity level of radiologists in this dataset, this system produced 19 less false-positives than the number of biopsied benign lesions in our dataset.

CONCLUSIONS

Use of adjunct imaging and PI has a significant contribution in diagnostic performance of ultrafast breast MRI. The developed AI system for interpretation of multiparametric ultrafast breast MRI may improve specificity.

Can apparent diffusion coefficient (ADC) distinguish breast cancer from benign breast findings? A meta-analysis based on 13 847 lesions

BACKGROUND

The purpose of the present meta-analysis was to provide evident data about use of Apparent Diffusion Coefficient (ADC) values for distinguishing malignant and benign breast lesions.

METHODS

MEDLINE library and SCOPUS database were screened for associations between ADC and malignancy/benignancy of breast lesions up to December 2018. Overall, 123 items were identified. The following data were extracted from the literature: authors, year of publication, study design, number of patients/lesions, lesion type, mean value and standard deviation of ADC, measure method, b values, and Tesla strength. The methodological quality of the 123 studies was checked according to the QUADAS-2 instrument. The meta-analysis was undertaken by using RevMan 5.3 software. DerSimonian and Laird random-effects models with inverse-variance weights were used without any further correction to account for the heterogeneity between the studies. Mean ADC values including 95% confidence intervals were calculated separately for benign and malign lesions.

RESULTS

The acquired 123 studies comprised 13,847 breast lesions. Malignant lesions were diagnosed in 10,622 cases (76.7%) and benign lesions in 3225 cases (23.3%). The mean ADC value of the malignant lesions was 1.03 × 10^- 3 mm²/s and the mean value of the benign lesions was 1.5 × 10^- 3 mm²/s. The calculated ADC values of benign lesions were over the value of 1.00 × 10^- 3 mm²/s. This result was independent on Tesla strength, choice of b values, and measure methods (whole lesion measure vs estimation of ADC in a single area).

CONCLUSION

An ADC threshold of 1.00 × 10^- 3 mm²/s can be recommended for distinguishing breast cancers from benign lesions.

Diffusion MRI of the breast: Current status and future directions

Diffusion-weighted imaging (DWI) is increasingly being incorporated into routine breast MRI protocols in many institutions worldwide, and there are abundant breast DWI indications ranging from lesion detection and distinguishing malignant from benign tumors to assessing prognostic biomarkers of breast cancer and predicting treatment response. DWI has the potential to serve as a noncontrast MR screening method. Beyond apparent diffusion coefficient (ADC) mapping, which is a commonly used quantitative DWI measure, advanced DWI models such as intravoxel incoherent motion (IVIM), non-Gaussian diffusion MRI, and diffusion tensor imaging (DTI) are extensively exploited in this field, allowing the characterization of tissue perfusion and architecture and improving diagnostic accuracy without the use of contrast agents. This review will give a summary of the clinical literature along with future directions. Level of Evidence: 5 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2020;52:70-90.

Risk stratification of triple-negative breast cancer with core gene signatures associated with chemoresponse and prognosis

PURPOSE

Neoadjuvant chemotherapy studies have consistently reported a strong correlation between pathologic response and long-term outcome in triple-negative breast cancer (TNBC). We aimed to define minimal gene signatures for predicting chemoresponse by a three-step approach and to further develop a risk-stratification method of TNBC.

METHODS

The first step involved the detection of genes associated with resistance to docetaxel in eight TNBC cell lines, leading to identification of thousands of candidate genes. Through subsequent second and third step analyses with gene set enrichment analysis and survival analysis using public expression profiles, the candidate gene list was reduced to prognostic core gene signatures comprising ten or four genes.

RESULTS

The prognostic core gene signatures include three up-regulated (CEBPD, MMP20, and WLS) and seven down-regulated genes (ASF1A, ASPSCR1, CHAF1B, DNMT1, GINS2, GOLGA2P5, and SKA1). We further develop a simple risk-stratification method based on expression profiles of the core genes. Relative expression values of the up-regulated and down-regulated core genes were averaged into two scores, Up and Down scores, respectively; then samples were stratified by a diagonal line in a xy plot of the Up and Down scores. Based on this method, the patients were successfully divided into subgroups with distinct chemoresponse and prognosis. The prognostic power of the method was validated in three independent public datasets containing 230, 141, and 117 TNBC patients with chemotherapy. In multivariable Cox regression analysis, the core gene signatures were significantly associated with prognosis independent of tumor stage and age at diagnosis. In meta-analysis, we found that five core genes (CEBPD, WLS, CHAF1B, GINS2, and SKA1) play opposing roles, either tumor promoter or suppressor, in TNBC and non-TNBC tumors respectively, depending on estrogen receptor status.

CONCLUSIONS

The results may provide a promising prognostic tool for predicting chemotherapy responders among TNBC patients prior to initiation of chemotherapeutic treatment.