Rationale and objectives

Breast segmentation using the U-net architecture was implemented and tested in independent validation datasets to quantify fibroglandular tissue volume in breast MRI.

Materials and methods

Two datasets were used. The training set was MRI of 286 patients with unilateral breast cancer. The segmentation was done on the contralateral normal breasts. The ground truth for the breast and fibroglandular tissue (FGT) was obtained by using a template-based segmentation method. The U-net deep learning algorithm was implemented to analyze the training set, and the final model was obtained using 10-fold cross-validation. The independent validation set was MRI of 28 normal volunteers acquired using four different MR scanners. Dice Similarity Coefficient (DSC), voxel-based accuracy, and Pearson's correlation were used to evaluate the performance.

Results

For the 10-fold cross-validation in the initial training set of 286 patients, the DSC range was 0.83-0.98 (mean 0.95 ± 0.02) for breast and 0.73-0.97 (mean 0.91 ± 0.03) for FGT; and the accuracy range was 0.92-0.99 (mean 0.98 ± 0.01) for breast and 0.87-0.99 (mean 0.97 ± 0.01) for FGT. For the entire 224 testing breasts of the 28 normal volunteers in the validation datasets, the mean DSC was 0.86 ± 0.05 for breast, 0.83 ± 0.06 for FGT; and the mean accuracy was 0.94 ± 0.03 for breast and 0.93 ± 0.04 for FGT. The testing results for MRI acquired using four different scanners were comparable.

Conclusion

Deep learning based on the U-net algorithm can achieve accurate segmentation results for the breast and FGT on MRI. It may provide a reliable and efficient method to process large number of MR images for quantitative analysis of breast density.

Background

Background parenchymal enhancement (BPE) on dynamic contrast-enhanced (DCE)-MRI has been associated with breast cancer risk, both based on qualitative and quantitative assessments.

Purpose

To investigate whether BPE of the contralateral breast on preoperative DCE-MRI is associated with therapy outcome in ER-positive, HER2-negative, node-negative invasive breast cancer.

Study type

Retrospective.

Population

In all, 289 patients with unilateral ER-positive, HER2-negative, node-negative breast cancer larger than 5 mm.

Field strength/sequence

3T, T₁ -weighted DCE sequence.

Assessment

BPE of the contralateral breast was assessed qualitatively by two dedicated radiologists and quantitatively (using region-of-interest and automatic breast segmentation).

Statistical tests

Cox regression analysis was used to determine associations with recurrence-free survival (RFS) and distant metastasis-free survival (DFS). Interobserver variability for parenchymal enhancement was assessed using kappa statistics and intraclass correlation coefficient (ICC).

Results

The median follow-up time was 75.8 months. Multivariate analysis showed receipt of total mastectomy (hazard ratio [HR]: 5.497) and high Ki-67 expression level (HR: 5.956) were independent factors associated with worse RFS (P < 0.05). Only a high Ki-67 expression level was associated with worse DFS (HR: 3.571, P = 0.045). BPE assessments were not associated with outcome (RFS [qualitative BPE: P = 0.75, 0.92 for readers 1 and 2; quantitative BPE: P = 0.38-0.99], DFS, [qualitative BPE: P = 0.41, 0.16 for readers 1 and 2; quantitative BPE: P = 0.68-0.99]). For interobserver variability, there was good agreement between qualitative (κ = 0.700) and good to perfect agreement for most quantitative parameters of BPE.

Data conclusion

Contralateral BPE showed no association with survival outcome in patients with ER-positive, HER2-negative, node-negative invasive breast cancer. A high Ki-67 expression level was associated with both worse recurrence-free and distant metastasis-free survival.

Level of evidence

3 Technical Efficacy: Stage 4 J. Magn. Reson. Imaging 2018;48:1678-1689.

Background

To correlate parameters of Ultrasonography-guided Diffuse optical tomography (US-DOT) with pharmacokinetic features of Dynamic contrast-enhanced (DCE)-MRI and pathologic markers of breast cancer.

Methods

Our institutional review board approved this retrospective study and waived the requirement for informed consent. Thirty seven breast cancer patients received US-DOT and DCE-MRI with less than two weeks in between imaging sessions. The maximal total hemoglobin concentration (THC) measured by US-DOT was correlated with DCE-MRI pharmacokinetic parameters, which included K(trans), k ep and signal enhancement ratio (SER). These imaging parameters were also correlated with the pathologic biomarkers of breast cancer.

Results

The parameters THC and SER showed marginal positive correlation (r = 0.303, p = 0.058). Tumors with high histological grade, negative ER, and higher Ki-67 expression ≥ 20% showed statistically higher THC values compared to their counterparts (p = 0.019, 0.041, and 0.023 respectively). Triple-negative (TN) breast cancers showed statistically higher K(trans) values than non-TN cancers (p = 0.048).

Conclusion

THC obtained from US-DOT and K(trans) obtained from DCE-MRI were associated with biomarkers indicative of a higher aggressiveness in breast cancer. Although US-DOT and DCE-MRI both measured the vascular properties of breast cancer, parameters from the two imaging modalities showed a weak association presumably due to their different contrast mechanisms and depth sensitivities.

Rationale and objectives

Computer-aided methods have been widely applied to diagnose lesions on breast magnetic resonance imaging (MRI). The first step was to identify abnormal areas. A deep learning Mask Regional Convolutional Neural Network (R-CNN) was implemented to search the entire set of images and detect suspicious lesions.

Materials and methods

Two DCE-MRI datasets were used, 241 patients acquired using non-fat-sat sequence for training, and 98 patients acquired using fat-sat sequence for testing. All patients have confirmed unilateral mass cancers. The tumor was segmented using fuzzy c-means clustering algorithm to serve as the ground truth. Mask R-CNN was implemented with ResNet-101 as the backbone. The neural network output the bounding boxes and the segmented tumor for evaluation using the Dice Similarity Coefficient (DSC). The detection performance, and the trade-off between sensitivity and specificity, was analyzed using free response receiver operating characteristic.

Results

When the precontrast and subtraction image of both breasts were used as input, the false positive from the heart and normal parenchymal enhancements could be minimized. The training set had 1469 positive slices (containing lesion) and 9135 negative slices. In 10-fold cross-validation, the mean accuracy = 0.86 and DSC = 0.82. The testing dataset had 1568 positive and 7264 negative slices, with accuracy = 0.75 and DSC = 0.79. When the obtained per-slice results were combined, 240 of 241 (99.5%) lesions in the training and 98 of 98 (100%) lesions in the testing datasets were identified.

Conclusion

Deep learning using Mask R-CNN provided a feasible method to search breast MRI, localize, and segment lesions. This may be integrated with other artificial intelligence algorithms to develop a fully automatic breast MRI diagnostic system.