Generalizable attention U-Net for segmentation of fibroglandular tissue and background parenchymal enhancement in breast DCE-MRI.

OBJECTIVES: Development of automated segmentation models enabling standardized volumetric quantification of fibroglandular tissue (FGT) from native volumes and background parenchymal enhancement (BPE) from subtraction volumes of dynamic contrast-enhanced breast MRI. Subsequent assessment of the developed models in the context of FGT and BPE Breast Imaging Reporting and Data System (BI-RADS)-compliant classification. METHODS: For the training and validation of attention U-Net models, data coming from a single 3.0-T scanner was used. For testing, additional data from 1.5-T scanner and data acquired in a different institution with a 3.0-T scanner was utilized. The developed models were used to quantify the amount of FGT and BPE in 80 DCE-MRI examinations, and a correlation between these volumetric measures and the classes assigned by radiologists was performed. RESULTS: To assess the model performance using application-relevant metrics, the correlation between the volumes of breast, FGT, and BPE calculated from ground truth masks and predicted masks was checked. Pearson correlation coefficients ranging from 0.963 ± 0.004 to 0.999 ± 0.001 were achieved. The Spearman correlation coefficient for the quantitative and qualitative assessment, i.e., classification by radiologist, of FGT amounted to 0.70 (p < 0.0001), whereas BPE amounted to 0.37 (p = 0.0006). CONCLUSIONS: Generalizable algorithms for FGT and BPE segmentation were developed and tested. Our results suggest that when assessing FGT, it is sufficient to use volumetric measures alone. However, for the evaluation of BPE, additional models considering voxels' intensity distribution and morphology are required. CRITICAL RELEVANCE STATEMENT: A standardized assessment of FGT density can rely on volumetric measures, whereas in the case of BPE, the volumetric measures constitute, along with voxels' intensity distribution and morphology, an important factor. KEY POINTS: • Our work contributes to the standardization of FGT and BPE assessment. • Attention U-Net can reliably segment intricately shaped FGT and BPE structures. • The developed models were robust to domain shift.

Automatic and standardized quality assurance of digital mammography and tomosynthesis with deep convolutional neural networks.

OBJECTIVES: The aim of this study was to develop and validate a commercially available AI platform for the automatic determination of image quality in mammography and tomosynthesis considering a standardized set of features. MATERIALS AND METHODS: In this retrospective study, 11,733 mammograms and synthetic 2D reconstructions from tomosynthesis of 4200 patients from two institutions were analyzed by assessing the presence of seven features which impact image quality in regard to breast positioning. Deep learning was applied to train five dCNN models on features detecting the presence of anatomical landmarks and three dCNN models for localization features. The validity of models was assessed by the calculation of the mean squared error in a test dataset and was compared to the reading by experienced radiologists. RESULTS: Accuracies of the dCNN models ranged between 93.0% for the nipple visualization and 98.5% for the depiction of the pectoralis muscle in the CC view. Calculations based on regression models allow for precise measurements of distances and angles of breast positioning on mammograms and synthetic 2D reconstructions from tomosynthesis. All models showed almost perfect agreement compared to human reading with Cohen's kappa scores above 0.9. CONCLUSIONS: An AI-based quality assessment system using a dCNN allows for precise, consistent and observer-independent rating of digital mammography and synthetic 2D reconstructions from tomosynthesis. Automation and standardization of quality assessment enable real-time feedback to technicians and radiologists that shall reduce a number of inadequate examinations according to PGMI (Perfect, Good, Moderate, Inadequate) criteria, reduce a number of recalls and provide a dependable training platform for inexperienced technicians.

Diagnostic accuracy of automated ACR BI-RADS breast density classification using deep convolutional neural networks.

OBJECTIVES: High breast density is a well-known risk factor for breast cancer. This study aimed to develop and adapt two (MLO, CC) deep convolutional neural networks (DCNN) for automatic breast density classification on synthetic 2D tomosynthesis reconstructions. METHODS: In total, 4605 synthetic 2D images (1665 patients, age: 57 ± 37 years) were labeled according to the ACR (American College of Radiology) density (A-D). Two DCNNs with 11 convolutional layers and 3 fully connected layers each, were trained with 70% of the data, whereas 20% was used for validation. The remaining 10% were used as a separate test dataset with 460 images (380 patients). All mammograms in the test dataset were read blinded by two radiologists (reader 1 with two and reader 2 with 11 years of dedicated mammographic experience in breast imaging), and the consensus was formed as the reference standard. The inter- and intra-reader reliabilities were assessed by calculating Cohen's kappa coefficients, and diagnostic accuracy measures of automated classification were evaluated. RESULTS: The two models for MLO and CC projections had a mean sensitivity of 80.4% (95%-CI 72.2-86.9), a specificity of 89.3% (95%-CI 85.4-92.3), and an accuracy of 89.6% (95%-CI 88.1-90.9) in the differentiation between ACR A/B and ACR C/D. DCNN versus human and inter-reader agreement were both "substantial" (Cohen's kappa: 0.61 versus 0.63). CONCLUSION: The DCNN allows accurate, standardized, and observer-independent classification of breast density based on the ACR BI-RADS system. KEY POINTS: • A DCNN performs on par with human experts in breast density assessment for synthetic 2D tomosynthesis reconstructions. • The proposed technique may be useful for accurate, standardized, and observer-independent breast density evaluation of tomosynthesis.

Detection of microcalcifications in photon-counting dedicated breast-CT using a deep convolutional neural network: Proof of principle.

OBJECTIVE: In this study, we investigate the feasibility of a deep Convolutional Neural Network (dCNN), trained with mammographic images, to detect and classify microcalcifications (MC) in breast-CT (BCT) images. METHODS: This retrospective single-center study was approved by the local ethics committee. 3518 icons generated from 319 mammograms were classified into three classes: "no MC" (1121), "probably benign MC" (1332), and "suspicious MC" (1065). A dCNN was trained (70% of data), validated (20%), and tested on a "real-world" dataset (10%). The diagnostic performance of the dCNN was tested on a subset of 60 icons, generated from 30 mammograms and 30 breast-CT images, and compared to human reading. ROC analysis was used to calculate diagnostic performance. Moreover, colored probability maps for representative BCT images were calculated using a sliding-window approach. RESULTS: The dCNN reached an accuracy of 98.8% on the "real-world" dataset. The accuracy on the subset of 60 icons was 100% for mammographic images, 60% for "no MC", 80% for "probably benign MC" and 100% for "suspicious MC". Intra-class correlation between the dCNN and the readers was almost perfect (0.85). Kappa values between the two readers (0.93) and the dCNN were almost perfect (reader 1: 0.85 and reader 2: 0.82). The sliding-window approach successfully detected suspicious MC with high image quality. The diagnostic performance of the dCNN to classify benign and suspicious MC was excellent with an AUC of 93.8% (95% CI 87, 4%-100%). CONCLUSION: Deep convolutional networks can be used to detect and classify benign and suspicious MC in breast-CT images.

BI-RADS-Based Classification of Mammographic Soft Tissue Opacities Using a Deep Convolutional Neural Network.

The aim of this study was to investigate the potential of a machine learning algorithm to classify breast cancer solely by the presence of soft tissue opacities in mammograms, independent of other morphological features, using a deep convolutional neural network (dCNN). Soft tissue opacities were classified based on their radiological appearance using the ACR BI-RADS atlas. We included 1744 mammograms from 438 patients to create 7242 icons by manual labeling. The icons were sorted into three categories: "no opacities" (BI-RADS 1), "probably benign opacities" (BI-RADS 2/3) and "suspicious opacities" (BI-RADS 4/5). A dCNN was trained (70% of data), validated (20%) and finally tested (10%). A sliding window approach was applied to create colored probability maps for visual impression. Diagnostic performance of the dCNN was compared to human readout by experienced radiologists on a "real-world" dataset. The accuracies of the models on the test dataset ranged between 73.8% and 89.8%. Compared to human readout, our dCNN achieved a higher specificity (100%, 95% CI: 85.4-100%; reader 1: 86.2%, 95% CI: 67.4-95.5%; reader 2: 79.3%, 95% CI: 59.7-91.3%), and the sensitivity (84.0%, 95% CI: 63.9-95.5%) was lower than that of human readers (reader 1:88.0%, 95% CI: 67.4-95.4%; reader 2:88.0%, 95% CI: 67.7-96.8%). In conclusion, a dCNN can be used for the automatic detection as well as the standardized and observer-independent classification of soft tissue opacities in mammograms independent of the presence of microcalcifications. Human decision making in accordance with the BI-RADS classification can be mimicked by artificial intelligence.

Radiomics in photon-counting dedicated breast CT: potential of texture analysis for breast density classification.

BACKGROUND: We investigated whether features derived from texture analysis (TA) can distinguish breast density (BD) in spiral photon-counting breast computed tomography (PC-BCT). METHODS: In this retrospective single-centre study, we analysed 10,000 images from 400 PC-BCT examinations of 200 patients. Images were categorised into four-level density scale (a-d) using Breast Imaging Reporting and Data System (BI-RADS)-like criteria. After manual definition of representative regions of interest, 19 texture features (TFs) were calculated to analyse the voxel grey-level distribution in the included image area. ANOVA, cluster analysis, and multinomial logistic regression statistics were used. A human readout then was performed on a subset of 60 images to evaluate the reliability of the proposed feature set. RESULTS: Of the 19 TFs, 4 first-order features and 7 second-order features showed significant correlation with BD and were selected for further analysis. Multinomial logistic regression revealed an overall accuracy of 80% for BD assessment. The majority of TFs systematically increased or decreased with BD. Skewness (rho -0.81), as a first-order feature, and grey-level nonuniformity (GLN, -0.59), as a second-order feature, showed the strongest correlation with BD, independently of other TFs. Mean skewness and GLN decreased linearly from density a to d. Run-length nonuniformity (RLN), as a second-order feature, showed moderate correlation with BD, but resulted in redundant being correlated with GLN. All other TFs showed only weak correlation with BD (range -0.49 to 0.49, p < 0.001) and were neglected. CONCLUSION: TA of PC-BCT images might be a useful approach to assess BD and may serve as an observer-independent tool.

Detecting Abnormal Axillary Lymph Nodes on Mammograms Using a Deep Convolutional Neural Network.

The purpose of this study was to determine the feasibility of a deep convolutional neural network (dCNN) to accurately detect abnormal axillary lymph nodes on mammograms. In this retrospective study, 107 mammographic images in mediolateral oblique projection from 74 patients were labeled to three classes: (1) "breast tissue", (2) "benign lymph nodes", and (3) "suspicious lymph nodes". Following data preprocessing, a dCNN model was trained and validated with 5385 images. Subsequently, the trained dCNN was tested on a "real-world" dataset and the performance compared to human readers. For visualization, colored probability maps of the classification were calculated using a sliding window approach. The accuracy was 98% for the training and 99% for the validation set. Confusion matrices of the "real-world" dataset for the three classes with radiological reports as ground truth yielded an accuracy of 98.51% for breast tissue, 98.63% for benign lymph nodes, and 95.96% for suspicious lymph nodes. Intraclass correlation of the dCNN and the readers was excellent (0.98), and Kappa values were nearly perfect (0.93-0.97). The colormaps successfully detected abnormal lymph nodes with excellent image quality. In this proof-of-principle study in a small patient cohort from a single institution, we found that deep convolutional networks can be trained with high accuracy and reliability to detect abnormal axillary lymph nodes on mammograms.

Fully automatic classification of automated breast ultrasound (ABUS) imaging according to BI-RADS using a deep convolutional neural network.

PURPOSE: The aim of this study was to develop and test a post-processing technique for detection and classification of lesions according to the BI-RADS atlas in automated breast ultrasound (ABUS) based on deep convolutional neural networks (dCNNs). METHODS AND MATERIALS: In this retrospective study, 645 ABUS datasets from 113 patients were included; 55 patients had lesions classified as high malignancy probability. Lesions were categorized in BI-RADS 2 (no suspicion of malignancy), BI-RADS 3 (probability of malignancy < 3%), and BI-RADS 4/5 (probability of malignancy > 3%). A deep convolutional neural network was trained after data augmentation with images of lesions and normal breast tissue, and a sliding-window approach for lesion detection was implemented. The algorithm was applied to a test dataset containing 128 images and performance was compared with readings of 2 experienced radiologists. RESULTS: Results of calculations performed on single images showed accuracy of 79.7% and AUC of 0.91 [95% CI: 0.85-0.96] in categorization according to BI-RADS. Moderate agreement between dCNN and ground truth has been achieved (κ: 0.57 [95% CI: 0.50-0.64]) what is comparable with human readers. Analysis of whole dataset improved categorization accuracy to 90.9% and AUC of 0.91 [95% CI: 0.77-1.00], while achieving almost perfect agreement with ground truth (κ: 0.82 [95% CI: 0.69-0.95]), performing on par with human readers. Furthermore, the object localization technique allowed the detection of lesion position slice-wise. CONCLUSIONS: Our results show that a dCNN can be trained to detect and distinguish lesions in ABUS according to the BI-RADS classification with similar accuracy as experienced radiologists. KEY POINTS: • A deep convolutional neural network (dCNN) was trained for classification of ABUS lesions according to the BI-RADS atlas. • A sliding-window approach allows accurate automatic detection and classification of lesions in ABUS examinations.

Applied Machine Learning in Spiral Breast-CT: Can We Train a Deep Convolutional Neural Network for Automatic, Standardized and Observer Independent Classification of Breast Density?

The aim of this study was to investigate the potential of a machine learning algorithm to accurately classify parenchymal density in spiral breast-CT (BCT), using a deep convolutional neural network (dCNN). In this retrospectively designed study, 634 examinations of 317 patients were included. After image selection and preparation, 5589 images from 634 different BCT examinations were sorted by a four-level density scale, ranging from A to D, using ACR BI-RADS-like criteria. Subsequently four different dCNN models (differences in optimizer and spatial resolution) were trained (70% of data), validated (20%) and tested on a "real-world" dataset (10%). Moreover, dCNN accuracy was compared to a human readout. The overall performance of the model with lowest resolution of input data was highest, reaching an accuracy on the "real-world" dataset of 85.8%. The intra-class correlation of the dCNN and the two readers was almost perfect (0.92) and kappa values between both readers and the dCNN were substantial (0.71-0.76). Moreover, the diagnostic performance between the readers and the dCNN showed very good correspondence with an AUC of 0.89. Artificial Intelligence in the form of a dCNN can be used for standardized, observer-independent and reliable classification of parenchymal density in a BCT examination.

Fully automatic classification of breast MRI background parenchymal enhancement using a transfer learning approach.

Marked enhancement of the fibroglandular tissue on contrast-enhanced breast magnetic resonance imaging (MRI) may affect lesion detection and classification and is suggested to be associated with higher risk of developing breast cancer. The background parenchymal enhancement (BPE) is qualitatively classified according to the BI-RADS atlas into the categories "minimal," "mild," "moderate," and "marked." The purpose of this study was to train a deep convolutional neural network (dCNN) for standardized and automatic classification of BPE categories.This IRB-approved retrospective study included 11,769 single MR images from 149 patients. The MR images were derived from the subtraction between the first post-contrast volume and the native T1-weighted images. A hierarchic approach was implemented relying on 2 dCNN models for detection of MR-slices imaging breast tissue and for BPE classification, respectively. Data annotation was performed by 2 board-certified radiologists. The consensus of the 2 radiologists was chosen as reference for BPE classification. The clinical performances of the single readers and of the dCNN were statistically compared using the quadratic Cohen's kappa.Slices depicting the breast were classified with training, validation, and real-world (test) accuracies of 98%, 96%, and 97%, respectively. Over the 4 classes, the BPE classification was reached with mean accuracies of 74% for training, 75% for the validation, and 75% for the real word dataset. As compared to the reference, the inter-reader reliabilities for the radiologists were 0.780 (reader 1) and 0.679 (reader 2). On the other hand, the reliability for the dCNN model was 0.815.Automatic classification of BPE can be performed with high accuracy and support the standardization of tissue classification in MRI.

Assessment of the systematic errors caused by diffusion gradient inhomogeneity in DTI-computer simulations.

Diffusion tensor imaging (DTI) is a powerful MRI modality that allows the investigation of the microstructure of tissues both in vivo and noninvasively. Its reliability is strictly dependent on the performance of diffusion-sensitizing gradients, of which spatial nonuniformity is a known issue in the case of virtually all clinical MRI scanners. The influence of diffusion gradient inhomogeneity on the accuracy of the diffusion tensor imaging was investigated by means of computer simulations supported by an MRI experiment performed at the isocenter and 15 cm away. The DTI measurements of two diffusion phantoms were simulated assuming a nonuniform diffusion-sensitizing gradient and various levels of noise. Thereafter, the tensors were calculated by two methods: (i) assuming a spatially constant b-matrix (standard DTI) and (ii) applying the b-matrix spatial distribution in the DTI (BSD-DTI) technique, a method of indicating the b-matrix for each voxel separately using an anisotropic phantom as a standard of diffusion. The average eigenvalues and fractional anisotropy across the homogeneous region of interest were calculated and compared with the expected values. Diffusion gradient inhomogeneity leads to overestimation of the largest eigenvalue, underestimation of the smallest one and thus overestimation of fractional anisotropy. The effect is similar to that caused by noise; however, it could not be corrected by increasing SNR. The MRI measurements, performed using a 3 T clinical scanner, revealed that the split of the eigenvalues measured 15 cm away from the isocenter is significant (up to 25%). The BSD-DTI calibration allowed the reduction of the measured fractional anisotropy of the isotropic medium from 0.174 to 0.031, suggesting that gradient inhomogeneity was the main cause of this error. For the phantom measured at the isocenter, however, the split was almost not observed; the average eigenvalues were shifted from the expected value by ~ 5%.

Analysis and correction of errors in DTI-based tractography due to diffusion gradient inhomogeneity.

The DTI-based tractography, despite its restrictions, is the most widely utilized fiber tracking method in clinical practice. Its fidelity is strictly dependent on the precision and accuracy of the DTI measurement, which in turn is limited by the linearity of the diffusion sensitizing gradient. The influence of the gradient distortions on the differences between the real and measured orientation of fibers was investigated by computer simulations. In addition, the potential of the b-matrix Spatial Distribution in DTI (BSD-DTI) technique in correcting such kind of errors was demonstrated experimentally. The simulations revealed that the diffusion gradient inhomogeneity, if not corrected, leads to the erroneous indication of the fiber direction. The average and maximum deviations were about 1° and 15°, respectively. Remarkably, the deviation between the real and measured orientation of fibers is directionally dependent, what was confirmed in MRI measurement. The deviation errors can be effectively corrected by preceding the DTI measurement with the BSD-DTI calibration.

The generalized Stejskal-Tanner equation for non-uniform magnetic field gradients.

The intensity of the diffusion weighted NMR signal is described by the Stejskal-Tanner equation, which was derived under the assumption that the gradients are uniform throughout the sample. Nevertheless, it has been demonstrated numerous times that this condition is not fulfilled in the cases of virtually any clinical or research MRI scanners. Therefore, technically, the Stejskal-Tanner equation is valid only for a very specific case of homogeneous gradients. In this paper the Stejskal-Tanner equation was derived for the general case on non-uniform diffusion gradients. To this end, the magnetic field was expressed as linear in a curvilinear coordinate system defined by a vector function p(r). Thereafter, the expression for the non-linear magnetic field was put into the Bloch-Torrey equation and solved. Moreover, the meaning of so-called coil tensor, which is used for the gradients inhomogeneity correction, was explained. It was proven that in the case of the spin echo-based sequences, the Stejskal-Tenner equation is still valid, even if the diffusion gradients are non-uniform. However, in such a case, the b-matrix should be derived for each voxel separately. For other sequence, the derived relation possesses an imaginary term, which corresponds do the phase shift of the diffusion weighted signal.

A theoretical validation of the B-matrix spatial distribution approach to diffusion tensor imaging.

The recently presented B-matrix Spatial Distribution (BSD) approach is a calibration technique which derives the actual distribution of the B-matrix in space. It is claimed that taking into account the spatial variability of the B-matrix improves the accuracy of diffusion tensor imaging (DTI). The purpose of this study is to verify this approach theoretically through computer simulations. Assuming three different spatial distributions of the B-matrix, diffusion weighted signals were calculated for the six orientations of a model anisotropic phantom. Subsequently two variants of the BSD calibration were performed for each of the three cases; one with the assumption of high uniformity of the model phantom (uBSD-DTI) and the other taking into account imperfections in phantom structure (BSD-DTI). Several cases of varying degrees of phantom uniformity were analyzed and the distributions of the B-matrix obtained were used for the calculation of the diffusion tensor of a model isotropic phantom. The results were compared with standard diffusion tensor calculation. The simulations confirmed the improvement of accuracy in the determination of the diffusion tensor after the calibration. BSD-DTI improves accuracy independent of both the degree of uniformity of the phantom and the inhomogeneity of the B-matrix. In cases of a relatively good uniformity of the phantom and minor distortions in the spatial distribution of the B-matrix, the uBSD-DTI approach is sufficient.

Theoretical analysis of phantom rotations in BSD-DTI.

A novel method of improving accuracy of diffusion tensor imaging (DTI), called BSD-DTI (B-spatial distribution in DTI), has been recently proposed. Determination of the b matrix components using an anisotropic phantom, and derivation of the spatial distribution are of the essence in this approach. So far, a sufficient uniformity of the diffusion properties across the entire phantom has been assumed. Nevertheless, BSD-DTI is not limited only to highly homogeneous phantoms. This study describes a procedure which allows to use basically any anisotropic phantom of a precisely defined structure.