Quantitative 23 Na-MRI of the intervertebral disk at 3 T.

Monitoring the tissue sodium content (TSC) in the intervertebral disk geometry noninvasively by MRI is a sensitive measure to estimate changes in the proteoglycan content of the intervertebral disk, which is a biomarker of degenerative disk disease (DDD) and of lumbar back pain (LBP). However, application of quantitative sodium concentration measurements in 23 Na-MRI is highly challenging due to the lower in vivo concentrations and smaller gyromagnetic ratio, ultimately yielding much smaller signal relative to 1 H-MRI. Moreover, imaging the intervertebral disk geometry imposes higher demands, mainly because the necessary RF volume coils produce highly inhomogeneous transmit field patterns. For an accurate absolute quantification of TSC in the intervertebral disks, the B1 field variations have to be mitigated. In this study, we report for the first time quantitative sodium concentration in the intervertebral disks at clinical field strengths (3 T) by deploying 23 Na-MRI in healthy human subjects. The sodium B1 maps were calculated by using the double-angle method and a double-tuned (1 H/23 Na) transceive chest coil, and the individual effects of the variation in the B1 field patterns in tissue sodium quantification were calculated. Phantom measurements were conducted to evaluate the quality of the Na-weighted images and B1 mapping. Depending on the disk position, the sodium concentration was calculated as 161.6 mmol/L-347 mmol/L, and the mean sodium concentration of the intervertebral disks varies between 254.6 ± 54 mmol/L and 290.1 ± 39 mmol/L. A smoothing effect of the B1 correction on the sodium concentration maps was observed, such that the standard deviation of the mean sodium concentration was significantly reduced with B1 mitigation. The results of this work provide an improved integration of quantitative 23 Na-MRI into clinical studies in intervertebral disks such as degenerative disk disease and establish alternative scoring schemes to existing morphological scoring such as the Pfirrmann score.

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.

23 Na-T1 quantification with saturation recovery TrueFISP and variable flip angle GRE at 3T: A phantom study.

PURPOSE: The aim of the current study was to compare the reproducibility of sodium (23 Na)-T1 estimation using a centric-reordered saturation recovery (SR) true fast imaging with steady-state precession (TrueFISP) and a variable flip angle (VFA) spoiled gradient echo (GRE). Additionally, we evaluated the effect of spatial averaging on 23 Na-T1 estimation by the two methods. METHODS: Measurements were performed in the phantom, consisting of 10 dm3 volume rectangular polyethylene container filled with distilled water solution of 0.6% NaCl + 0.004% CuSO4 , using a dual-tunable 23 Na/1 H coil at 3 Tesla. 23 Na images were acquired for FOV = 384 × 384 mm2 and voxel size = 6 × 6 × 6 mm3 using: (1) TrueFISP: TR/TE = 900/1.5 ms, flip angle = 90°, bandwidth = 450 Hz/px, and (2) GRE: TR/TE = 30/1.5 ms, bandwidth = 350 Hz/px. 23 Na-T1 weightings were obtained with nonselective saturation prepulses delayed from the center of the k-space acquisition by 25/40/60/130/280 ms (SR-TrueFISP) and by applying different nominal flip angles: 10°/30°/50°/70°/90° (VFA-GRE). Both sequences were acquired twice, applying 20 and 30 spatial averages. The resulting images were B1 -corrected with a double-angle GRE method. RESULTS: Image acquisition varied from 5:41 to 9:37 for TrueFISP and from 12:48 to 19:12 min for GRE using 20 and 30 spatial averages, respectively. Higher averaging increased the acquisition time by 53% and mean SNR at scan < 10%, without an effect on 23 Na-T1 estimations with both methods (SR-Truefisp |Δ| = 1.58 ms, VFA-GRE |Δ| = 0.53 ms; for SNR P < .001). Overall, mean ± SD of 23 Na-T1 was found as 51 ± 3 ms with SR-TrueFISP and 53 ± 2 ms with VFA-GRE. CONCLUSION: Both SR-TrueFISP and VFA-GRE provided similar 23 Na-T1 estimates based on the phantom measurements with isotropic resolution.

Diagnostic accuracy of texture analysis and machine learning for quantification of liver fibrosis in MRI: correlation with MR elastography and histopathology.

OBJECTIVES: To compare the diagnostic accuracy of texture analysis (TA)-derived parameters combined with machine learning (ML) of non-contrast-enhanced T1w and T2w fat-saturated (fs) images with MR elastography (MRE) for liver fibrosis quantification. METHODS: In this IRB-approved prospective study, liver MRIs of participants with suspected chronic liver disease who underwent liver biopsy between August 2015 and May 2018 were analyzed. Two readers blinded to clinical and histopathological findings performed TA. The participants were categorized into no or low-stage (0-2) and high-stage (3-4) fibrosis groups. Confusion matrices were calculated using a support vector machine combined with principal component analysis. The diagnostic accuracy of ML-based TA of liver fibrosis and MRE was assessed by area under the receiver operating characteristic curves (AUC). Histopathology served as reference standard. RESULTS: A total of 62 consecutive participants (40 men; mean age ± standard deviation, 48 ± 13 years) were included. The accuracy of TA and ML on T1w was 85.7% (95% confidence interval [CI] 63.7-97.0) and 61.9% (95% CI 38.4-81.9) on T2w fs for classification of liver fibrosis into low-stage and high-stage fibrosis. The AUC for TA on T1w was similar to MRE (0.82 [95% CI 0.59-0.95] vs. 0.92 [95% CI 0.71-0.99], p = 0.41), while the AUC for T2w fs was significantly lower compared to MRE (0.57 [95% CI 0.34-0.78] vs. 0.92 [95% CI 0.71-0.99], p = 0.008). CONCLUSION: Our results suggest that liver fibrosis can be quantified with TA-derived parameters of T1w when combined with a ML algorithm with similar accuracy compared to MRE. KEY POINTS: • Liver fibrosis can be categorized into low-stage fibrosis (0-2) and high-stage fibrosis (3-4) using texture analysis-derived parameters of T1-weighted images with a machine learning approach. • For the differentiation of low-stage fibrosis and high-stage fibrosis, the diagnostic accuracy of texture analysis on T1-weighted images combined with a machine learning algorithm is similar compared to MR elastography.

Quantification of sodium T1 in abdominal tissues at 3 T.

INTRODUCTION: Although relevant for assessment of sodium in multiple endocrine pathways, 23Na-T1 quantification is challenging due to technical limitations (SAR, B1 inhomogeneity) or influence of tissue's local molecular dynamics. Hereby, we propose T1 quantification of 23Na-MRI signal acquired over the abdomen using a centric-reordered saturation-recovery (SR) true fast imaging with steady state precession (TrueFISP) sequence. MATERIALS AND METHODS: Measurements were performed at 3T using a dual-tunable 23Na/1H coil in 7 healthy volunteers (TR/TE = 858-928/1.57 ms; flip angle = 90°; bandwidth = 450 Hz/px; voxel size = 5 × 5 × 10 mm3). Variable T1-weighting was achieved applying non-selective saturation pre-pulses delayed from the centre of the k-space acquisition by 25, 40, 60, 120 and 250 ms. T1-curve fitting was performed slice-wise, separately for average intensity values from the manually segmented areas of the renal parenchyma and spinal canal, over the increasing SR times- assuming monoexponential signal pattern. RESULTS: Mean ± standard deviation of 23Na-T1 was found as 29 ± 10 ms and 35 ± 8 ms for the renal parenchyma and the spinal canal, respectively. DISCUSSION: 23Na-T1 quantification using a SR-TrueFISP is feasible in clinical settings, in the images constrained by clinically applicable acquisition time of reduced spatial resolution or averages.

Diffusion tensor imaging of the abdominal organs: Influence of oriented intravoxel flow compartments.

Water flow in partially oriented intravoxel compartments mimics an anisotropic fast-diffusion regime, which contributes to the signal attenuation in diffusion-weighted images. In the abdominal organs, this flow may reflect physiological fluid movements (eg, tubular urine flow in kidneys, or bile flow through the liver) and have a clinical relevance. This study investigated the influence of anisotropic intravoxel water flow on diffusion tensor imaging (DTI) of the abdominal organs. Diffusion-weighted images were acquired in five healthy volunteers using an EPI sequence with diffusion preparation (TR/TE: 1000 ms/71 ms; b-values: 0, 10, 20, 40, 70, 120, 250, 450, 700, 1000 s/mm2 ; 12 noncollinear diffusion-encoding directions). DTI of liver and kidneys was performed assuming (i) monoexponential decay of the diffusion-weighted signal, and (ii) accounting for potential anisotropy of the fast-diffusion compartments using a tensorial generalization of the IVIM model. Additionally, potential dependency of the metrics of the tensors from the anatomical location was evaluated. Significant differences in the metrics of the diffusion tensor (DT) were found in both liver and kidneys when comparing the two models. In both organs, the trace and the fractional anisotropy of the DT were significantly higher in the monoexponential model than when accounting for perfusion. The comparison of areas of the liver proximal to the hilum with distal regions and of renal cortex with the medulla also proved a location dependency of the size of the fast-diffusion compartments. Pseudo-diffusion correction in DTI enables the assessment of the solid parenchyma regardless of the organ perfusion or other pseudo-diffusive fluid movements. This may have a clinical relevance in the assessment of parenchymal pathologies (eg, liver fibrosis). The fast pseudo-diffusion components present a detectable anisotropy, which may reflect the hepatic microcirculation or other sources of mesoscopic fluid movement in the abdominal organs.

Automatic classification of ultrasound breast lesions using a deep convolutional neural network mimicking human decision-making.

OBJECTIVES: To evaluate a deep convolutional neural network (dCNN) for detection, highlighting, and classification of ultrasound (US) breast lesions mimicking human decision-making according to the Breast Imaging Reporting and Data System (BI-RADS). METHODS AND MATERIALS: One thousand nineteen breast ultrasound images from 582 patients (age 56.3 ± 11.5 years) were linked to the corresponding radiological report. Lesions were categorized into the following classes: no tissue, normal breast tissue, BI-RADS 2 (cysts, lymph nodes), BI-RADS 3 (non-cystic mass), and BI-RADS 4-5 (suspicious). To test the accuracy of the dCNN, one internal dataset (101 images) and one external test dataset (43 images) were evaluated by the dCNN and two independent readers. Radiological reports, histopathological results, and follow-up examinations served as reference. The performances of the dCNN and the humans were quantified in terms of classification accuracies and receiver operating characteristic (ROC) curves. RESULTS: In the internal test dataset, the classification accuracy of the dCNN differentiating BI-RADS 2 from BI-RADS 3-5 lesions was 87.1% (external 93.0%) compared with that of human readers with 79.2 ± 1.9% (external 95.3 ± 2.3%). For the classification of BI-RADS 2-3 versus BI-RADS 4-5, the dCNN reached a classification accuracy of 93.1% (external 95.3%), whereas the classification accuracy of humans yielded 91.6 ± 5.4% (external 94.1 ± 1.2%). The AUC on the internal dataset was 83.8 (external 96.7) for the dCNN and 84.6 ± 2.3 (external 90.9 ± 2.9) for the humans. CONCLUSION: dCNNs may be used to mimic human decision-making in the evaluation of single US images of breast lesion according to the BI-RADS catalog. The technique reaches high accuracies and may serve for standardization of highly observer-dependent US assessment. KEY POINTS: • Deep convolutional neural networks could be used to classify US breast lesions. • The implemented dCNN with its sliding window approach reaches high accuracies in the classification of US breast lesions. • Deep convolutional neural networks may serve for standardization in US BI-RADS classification.

A New Model for MR Evaluation of Liver Function with Gadoxetic Acid, Including Both Uptake and Excretion.

OBJECTIVES: Most existing models that are in use to model hepatic function through assessment of hepatic gadoxetic acid enhancement kinetics do not consider quantitative measures of gadoxetic excretion. We developed a model that allows a simultaneous quantitation of uptake and excretion of liver specific contrast agents. The aim was to improve the assessment of hepatic synthetic function, and provide quantitative measures of hepatic excretion function. METHODS: Sixteen patients underwent dynamic T1-weighted turbo gradient echo imaging at 1.5 T prior and after bolus injection of gadoxetic acid at 0.1 ml/kg. DCE-images were obtained for 30 min after injection. A dual-inlet two-compartment model was then used to fit the measured liver signal values. Four tissue parameters (extracellular volume fraction, arterial flow fraction, uptake rate and excretion half-time) were extracted for each liver segment. RESULTS: The proposed model provided a good fit to acquired data. Mean values for arterial flow fraction (0.08+-0.04), extracellular volume (0.20±0.08) and uptake rate (4.02 ±1.32 /100 ml/min) were comparable to those obtained with the conventional model (0.08±0.05, 0.21±0.12, and 4.93±1.74), but exhibited significantly less variation and improved fit quality. CONCLUSIONS: The proposed model is more accurate than existing conventional models and provides an additional excretion parameter. KEY POINTS: • Models of hepatic contrast agent uptake can be extended to include excretion. • Including an additional excretion parameter improves accuracy of the model. • Standard diagnostic sequences can be extended to incorporate the model.

Automated pixel-wise brain tissue segmentation of diffusion-weighted images via machine learning.

The diffusion-weighted (DW) MR signal sampled over a wide range of b-values potentially allows for tissue differentiation in terms of cellularity, microstructure, perfusion, and T2 relaxivity. This study aimed to implement a machine learning algorithm for automatic brain tissue segmentation from DW-MRI datasets, and to determine the optimal sub-set of features for accurate segmentation. DWI was performed at 3 T in eight healthy volunteers using 15 b-values and 20 diffusion-encoding directions. The pixel-wise signal attenuation, as well as the trace and fractional anisotropy (FA) of the diffusion tensor, were used as features to train a support vector machine classifier for gray matter, white matter, and cerebrospinal fluid classes. The datasets of two volunteers were used for validation. For each subject, tissue classification was also performed on 3D T1 -weighted data sets with a probabilistic framework. Confusion matrices were generated for quantitative assessment of image classification accuracy in comparison with the reference method. DWI-based tissue segmentation resulted in an accuracy of 82.1% on the validation dataset and of 82.2% on the training dataset, excluding relevant model over-fitting. A mean Dice coefficient (DSC) of 0.79 ± 0.08 was found. About 50% of the classification performance was attributable to five features (i.e. signal measured at b-values of 5/10/500/1200 s/mm2 and the FA). This reduced set of features led to almost identical performances for the validation (82.2%) and the training (81.4%) datasets (DSC = 0.79 ± 0.08). Machine learning techniques applied to DWI data allow for accurate brain tissue segmentation based on both morphological and functional information.

Automatic Detection of Post-Operative Clips in Mammography Using a U-Net Convolutional Neural Network.

BACKGROUND: After breast conserving surgery (BCS), surgical clips indicate the tumor bed and, thereby, the most probable area for tumor relapse. The aim of this study was to investigate whether a U-Net-based deep convolutional neural network (dCNN) may be used to detect surgical clips in follow-up mammograms after BCS. METHODS: 884 mammograms and 517 tomosynthetic images depicting surgical clips and calcifications were manually segmented and classified. A U-Net-based segmentation network was trained with 922 images and validated with 394 images. An external test dataset consisting of 39 images was annotated by two radiologists with up to 7 years of experience in breast imaging. The network's performance was compared to that of human readers using accuracy and interrater agreement (Cohen's Kappa). RESULTS: The overall classification accuracy on the validation set after 45 epochs ranged between 88.2% and 92.6%, indicating that the model's performance is comparable to the decisions of a human reader. In 17.4% of cases, calcifications have been misclassified as post-operative clips. The interrater reliability of the model compared to the radiologists showed substantial agreement (κreader1 = 0.72, κreader2 = 0.78) while the readers compared to each other revealed a Cohen's Kappa of 0.84, thus showing near-perfect agreement. CONCLUSIONS: With this study, we show that surgery clips can adequately be identified by an AI technique. A potential application of the proposed technique is patient triage as well as the automatic exclusion of post-operative cases from PGMI (Perfect, Good, Moderate, Inadequate) evaluation, thus improving the quality management workflow.

Explainable Precision Medicine in Breast MRI: A Combined Radiomics and Deep Learning Approach for the Classification of Contrast Agent Uptake.

In DCE-MRI, the degree of contrast uptake in normal fibroglandular tissue, i.e., background parenchymal enhancement (BPE), is a crucial biomarker linked to breast cancer risk and treatment outcome. In accordance with the Breast Imaging Reporting & Data System (BI-RADS), it should be visually classified into four classes. The susceptibility of such an assessment to inter-reader variability highlights the urgent need for a standardized classification algorithm. In this retrospective study, the first post-contrast subtraction images for 27 healthy female subjects were included. The BPE was classified slice-wise by two expert radiologists. The extraction of radiomic features from segmented BPE was followed by dataset splitting and dimensionality reduction. The latent representations were then utilized as inputs to a deep neural network classifying BPE into BI-RADS classes. The network's predictions were elucidated at the radiomic feature level with Shapley values. The deep neural network achieved a BPE classification accuracy of 84 ± 2% (p-value < 0.00001). Most of the misclassifications involved adjacent classes. Different radiomic features were decisive for the prediction of each BPE class underlying the complexity of the decision boundaries. A highly precise and explainable pipeline for BPE classification was achieved without user- or algorithm-dependent radiomic feature selection.

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.

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.

Systematic analysis of changes in radiomics features during dynamic breast-MRI: Evaluation of specific biomarkers.

OBJECTIVES: In this retrospective, single-center study we investigate the changes of radiomics features during dynamic breast-MRI for healthy tissue compared to benign and malignant lesions. METHODS: 60 patients underwent breast-MRI using a dynamic 3D gradient-echo sequence. Changes of 34 texture features (TF) in 30 benign and 30 malignant lesions were calculated for 5 dynamic datasets and corresponding 4 subtraction datasets. Statistical analysis was performed with ANOVA, and systematic changes in features were described by linear and polynomial regression models. RESULTS: ANOVA revealed significant differences (p < 0.05) between normal tissue and lesions in 13 TF, compared to 9 TF between benign and malignant lesions. Most TF showed significant differences in early dynamic and subtraction datasets. TF associated with homogeneity were suitable to discriminate between healthy parenchyma and lesions, whereas run-length features were more suitable to discriminate between benign and malignant lesions. Run length nonuniformity (RLN) was the only feature able to distinguish between all three classes with an AUC of 88.3%. Characteristic changes were observed with a systematic increase or decrease for most TF with mostly polynomial behavior. Slopes showed earlier peaks in malignant lesions, compared to benign lesions. Mean values for the coefficient of determination were higher during subtraction sequences, compared to dynamic sequences (benign: 0.98 vs 0. 72; malignant: 0.94 vs 0.74). CONCLUSIONS: TF of breast lesions follow characteristic patterns during dynamic breast-MRI, distinguishing benign from malignant lesions. Early dynamic and subtraction datasets are particularly suitable for texture analysis in breast-MRI. Features associated with tissue homogeneity seem to be indicative of benign lesions.

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.

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.

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.

Deep learning for the standardized classification of Ki-67 in vulva carcinoma: A feasibility study.

BACKGROUND: The aim of this study is to demonstrate the feasibility of automatic classification of Ki-67 histological immunostainings in patients with squamous cell carcinoma of the vulva using a deep convolutional neural network (dCNN). MATERIAL AND METHODS: For evaluation of the dCNN, we used 55 well characterized squamous cell carcinomas of the vulva in a tissue microarray (TMA) format in this retrospective study. The tumor specimens were classified in 3 different categories C1 (0-2%), C2 (2-20%) and C3 (>20%), representing the relation of the number of KI-67 positive tumor cells to all cancer cells on the TMA spot. Representative areas of the spots were manually labeled by extracting images of 351 × 280 pixels. A dCNN with 13 convolutional layers was used for the evaluation. Two independent pathologists classified 45 labeled images in order to compare the dCNN's results to human readouts. RESULTS: Using a small labeled dataset with 1020 images with equal distribution among classes, the dCNN reached an accuracy of 90.9% (93%) for the training (validation) data. Applying a larger dataset with additional 1017 labeled images resulted in an accuracy of 96.1% (91.4%) for the training (validation) dataset. For the human readout, there were no significant differences between the pathologists and the dCNN in Ki-67 classification results. CONCLUSION: The dCNN is capable of a standardized classification of Ki-67 staining in vulva carcinoma; therefore, it may be suitable for quality control and standardization in the assessment of tumor grading.

Classification of Mammographic Breast Microcalcifications Using a Deep Convolutional Neural Network: A BI-RADS-Based Approach.

MATERIALS AND METHODS: Over 56,000 images of 268 mammograms from 94 patients were labeled to 3 classes according to the BI-RADS standard: "no microcalcifications" (BI-RADS 1), "probably benign microcalcifications" (BI-RADS 2/3), and "suspicious microcalcifications" (BI-RADS 4/5). Using the preprocessed images, a dCNN was trained and validated, generating 3 types of models: BI-RADS 4 cohort, BI-RADS 5 cohort, and BI-RADS 4 + 5 cohort. For the final validation of the trained dCNN models, a test data set consisting of 141 images of 51 mammograms from 26 patients labeled according to the corresponding BI-RADS classification from the radiological reports was applied. The performances of the dCNN models were evaluated, classifying each of the mammograms and computing the accuracy in comparison to the classification from the radiological reports. For visualization, probability maps of the classification were generated. RESULTS: The accuracy on the validation set after 130 epochs was 99.5% for the BI-RADS 4 cohort, 99.6% for the BI-RADS 5 cohort, and 98.1% for the BI-RADS 4 + 5 cohort. Confusion matrices of the "real-world" test data set for the 3 cohorts were generated where the radiological reports served as ground truth. The resulting accuracy was 39.0% for the BI-RADS 4 cohort, 80.9% for BI-RADS 5 cohort, and 76.6% for BI-RADS 4 + 5 cohort. The probability maps exhibited excellent image quality with correct classification of microcalcification distribution. CONCLUSIONS: The dCNNs can be trained to successfully classify microcalcifications on mammograms according to the BI-RADS classification system in order to act as a standardized quality control tool providing the expertise of a team of radiologists.