Automated pectoral muscle identification on MLO-view mammograms: Comparison of deep neural network to conventional computer vision

Deep learning-based breast region segmentation in raw and processed digital mammograms: generalization across views and vendors

Purpose

We developed a segmentation method suited for both raw (for processing) and processed (for presentation) digital mammograms (DMs) that is designed to generalize across images acquired with systems from different vendors and across the two standard screening views.

Approach

A U-Net was trained to segment mammograms into background, breast, and pectoral muscle. Eight different datasets, including two previously published public sets and six sets of DMs from as many different vendors, were used, totaling 322 screen film mammograms (SFMs) and 4251 DMs (2821 raw/processed pairs and 1430 only processed) from 1077 different women. Three experiments were done: first training on all SFM and processed images, second also including all raw images in training, and finally testing vendor generalization by leaving one dataset out at a time.

Results

The model trained on SFM and processed mammograms achieved a good overall performance regardless of projection and vendor, with a mean (±std. dev.) dice score of 0.96 ± 0.06 for all datasets combined. When raw images were included in training, the mean (±std. dev.) dice score for the raw images was 0.95 ± 0.05 and for the processed images was 0.96 ± 0.04 . Testing on a dataset with processed DMs from a vendor that was excluded from training resulted in a difference in mean dice varying between - 0.23 to + 0.02 from that of the fully trained model.

Conclusions

The proposed segmentation method yields accurate overall segmentation results for both raw and processed mammograms independent of view and vendor. The code and model weights are made available.

Breast glandularity and mean glandular dose assessment using a deep learning framework: Virtual patients study

PURPOSE

Breast dosimetry in mammography is an important aspect of radioprotection since women are exposed periodically to ionizing radiation due to breast cancer screening programs. Mean glandular dose (MGD) is the standard quantity employed for the establishment of dose reference levels in retrospective population studies. However, MGD calculations requires breast glandularity estimation. This work proposes a deep learning framework for volume glandular fraction (VGF) estimations based on mammography images, which in turn are converted to glandularity values for MGD calculations.

METHODS

208 virtual breast phantoms were generated and compressed computationally. The mammography images were obtained with Monte Carlo simulations (MC-GPU code) and a ray-tracing algorithm was employed for labeling the training data. The architectures of the neural networks are based on the XNet and multilayer perceptron, adapted for each task. The network predictions were compared with the ground truth using the coefficient of determination (r²).

RESULTS

The results have shown a good agreement for inner breast segmentation (r² = 0.999), breast volume prediction (r² = 0.982) and VGF prediction (r² = 0.935). Moreover, the DgN coefficients using the predicted VGF for the virtual population differ on average 1.3% from the ground truth values. Afterwards with the obtained DgN coefficients, the MGD values were estimated from exposure factors extracted from the DICOM header of a clinical cohort, with median(75 percentile) values of 1.91(2.45) mGy.

CONCLUSION

We successfully implemented a deep learning framework for VGF and MGD calculations for virtual breast phantoms.

Deep learning of mammary gland distribution for architectural distortion detection in digital breast tomosynthesis

Computer aided detection (CADe) for breast lesions can provide an important reference for radiologists in breast cancer screening. Architectural distortion (AD) is a type of breast lesion that is difficult to detect. A majority of CADe methods focus on detecting the radial pattern, which is a main characteristic of typical ADs. However, a few atypical ADs do not exhibit such a pattern. To improve the performance of CADe for typical and atypical ADs, we propose a deep-learning-based model that used mammary gland distribution as prior information to detect ADs in digital breast tomosynthesis (DBT). First, information about gland distribution, including the Gabor magnitude, the Gabor orientation field, and a convergence map, were produced using a bank of Gabor filters and convergence measures. Then, this prior information and an original slice were input into a Faster R-CNN detection network to obtain the 2-D candidates for each slice. Finally, a 3-D aggregation scheme was employed to fuse these 2-D candidates as 3-D candidates for each DBT volume. Retrospectively, 64 typical AD volumes, 74 atypical AD volumes, and 127 normal volumes were collected. Six-fold cross-validation and mean true positive fraction (MTPF) were used to evaluate the model. Compared to an existing convergence-based model, our proposed model achieved an MTPF of 0.53 ± 0.04, 0.61 ± 0.05, and 0.45 ± 0.04 for all DBT volumes, typical + normal volumes, and atypical + normal volumes, respectively. These results were significantly better than those of 0.36 ± 0.03, 0.46 ± 0.04, and 0.28 ± 0.04 for a convergence-based model (p ≪ 0.01). These results indicate that employing the prior information of gland distribution and a deep learning method can improve the performance of CADe for AD.

Automatic segmentation of the pectoral muscle based on boundary identification and shape prediction

The purpose of this work is to identify the pectoral muscle region in mediolateral oblique (MLO) view mammograms even when the boundary is blurred or obscured. The problem is decoupled into two subproblems in our study: identifying parts of boundaries with high confidence and predicting the overall shape of the pectoral muscle. Due to the similarity in intensity and texture between pectoral muscle and gland tissue, we trained a deep neural network to distinguish them in the first subproblem. The boundary with high confidence can be obtained according to the consistency of predictions from multiple converged models. For the shape prediction problem, a generative adversarial network (GAN) is used to learn mapping from a given identified region and the breast shape to the overall pectoral muscle shape. Our method is evaluated on a mammogram dataset including 633 MLO view mammograms collected from three different datacenters. We take U-Net as our baseline model and the dataset is divided into three groups according to the performance of U-Net for evaluation. In all three groups, U-Net achieves 80.1%, 92.9%, and 98.3% in the Dice similarity coefficient, respectively, and our method achieves 85.2%, 94.8%, and 98.1% in the Dice similarity coefficient, respectively. The experiment shows that our method effectively estimates the pectoral muscle boundary, even parts of boundaries that are difficult to detect, and greatly improves the performance of segmentation in this case.