A Survey of Publicly Available MRI Datasets for Potential Use in Artificial Intelligence Research.

Artificial intelligence (AI) has the potential to bring transformative improvements to the field of radiology; yet, there are barriers to widespread clinical adoption. One of the most important barriers has been access to large, well-annotated, widely representative medical image datasets, which can be used to accurately train AI programs. Creating such datasets requires time and expertise and runs into constraints around data security and interoperability, patient privacy, and appropriate data use. Recognizing these challenges, several institutions have started curating and providing publicly available, high-quality datasets that can be accessed by researchers to advance AI models. The purpose of this work was to review the publicly available MRI datasets that can be used for AI research in radiology. Despite being an emerging field, a simple internet search for open MRI datasets presents an overwhelming number of results. Therefore, we decided to create a survey of the major publicly accessible MRI datasets in different subfields of radiology (brain, body, and musculoskeletal), and list the most important features of value to the AI researcher. To complete this review, we searched for publicly available MRI datasets and assessed them based on several parameters (number of subjects, demographics, area of interest, technical features, and annotations). We reviewed 110 datasets across sub-fields with 1,686,245 subjects in 12 different areas of interest ranging from spine to cardiac. This review is meant to serve as a reference for researchers to help spur advancements in the field of AI for radiology. LEVEL OF EVIDENCE: Level 4 TECHNICAL EFFICACY: Stage 6.

AI Applications to Breast MRI: Today and Tomorrow.

In breast imaging, there is an unrelenting increase in the demand for breast imaging services, partly explained by continuous expanding imaging indications in breast diagnosis and treatment. As the human workforce providing these services is not growing at the same rate, the implementation of artificial intelligence (AI) in breast imaging has gained significant momentum to maximize workflow efficiency and increase productivity while concurrently improving diagnostic accuracy and patient outcomes. Thus far, the implementation of AI in breast imaging is at the most advanced stage with mammography and digital breast tomosynthesis techniques, followed by ultrasound, whereas the implementation of AI in breast magnetic resonance imaging (MRI) is not moving along as rapidly due to the complexity of MRI examinations and fewer available dataset. Nevertheless, there is persisting interest in AI-enhanced breast MRI applications, even as the use of and indications of breast MRI continue to expand. This review presents an overview of the basic concepts of AI imaging analysis and subsequently reviews the use cases for AI-enhanced MRI interpretation, that is, breast MRI triaging and lesion detection, lesion classification, prediction of treatment response, risk assessment, and image quality. Finally, it provides an outlook on the barriers and facilitators for the adoption of AI in breast MRI. LEVEL OF EVIDENCE: 5 TECHNICAL EFFICACY: Stage 6.

Assessment of Hypoxia in Breast Cancer: Emerging Functional MR Imaging and Spectroscopy Techniques and Clinical Applications.

Breast cancer is one of the most prevalent forms of cancer affecting women worldwide. Hypoxia, a condition characterized by insufficient oxygen supply in tumor tissues, is closely associated with tumor aggressiveness, resistance to therapy, and poor clinical outcomes. Accurate assessment of tumor hypoxia can guide treatment decisions, predict therapy response, and contribute to the development of targeted therapeutic interventions. Over the years, functional magnetic resonance imaging (fMRI) and magnetic resonance spectroscopy (MRS) techniques have emerged as promising noninvasive imaging options for evaluating hypoxia in cancer. Such techniques include blood oxygen level-dependent (BOLD) MRI, oxygen-enhanced MRI (OE) MRI, chemical exchange saturation transfer (CEST) MRI, and proton MRS (1H-MRS). These may help overcome the limitations of the routinely used dynamic contrast-enhanced (DCE) MRI and diffusion-weighted imaging (DWI) techniques, contributing to better diagnosis and understanding of the biological features of breast cancer. This review aims to provide a comprehensive overview of the emerging functional MRI and MRS techniques for assessing hypoxia in breast cancer, along with their evolving clinical applications. The integration of these techniques in clinical practice holds promising implications for breast cancer management. EVIDENCE LEVEL: 5 TECHNICAL EFFICACY: Stage 1.

Breast MRI in patients with implantable loop recorder: initial experience.

OBJECTIVES: To investigate the feasibility of breast MRI exams and guided biopsies in patients with an implantable loop recorder (ILR) as well as the impact ILRs may have on image interpretation. MATERIALS AND METHODS: This retrospective study examined breast MRIs of patients with ILR, from April 2008 to September 2022. Radiological reports and electronic medical records were reviewed for demographic characteristics, safety concerns, and imaging findings. MR images were analyzed and compared statistically for artifact quantification on the various pulse sequences. RESULTS: Overall, 40/82,778 (0.049%) MRIs during the study period included ILR. All MRIs were completed without early termination. No patient-related or device-related adverse events occurred. ILRs were most commonly located in the left lower-inner quadrant (64.6%). The main artifact was a signal intensity (SI) void in a dipole formation in the ILR bed with or without areas of peripheral high SI. Artifacts appeared greatest in the cranio-caudal axis (p < 0.001), followed by the anterior-posterior axis (p < 0.001), and then the right-left axis. High peripheral rim-like SI artifacts appeared on the post-contrast and subtracted T1-weighted images, mimicking suspicious enhancement. Artifacts were most prominent on diffusion-weighted (p < 0.001), followed by T2-weighted and T1-weighted images. In eight patients, suspicious findings were found on MRI, resulting in four additional malignant lesions. Of six patients with left breast cancer, the tumor was completely visible in five cases and partially obscured in one. CONCLUSION: Breast MRI is feasible and safe among patients with ILR and may provide a significant diagnostic value, albeit with localized, characteristic artifacts. CLINICAL RELEVANCE STATEMENT: Indicated breast MRI exams and guided biopsies can be safely performed in patients with implantable loop recorder. Nevertheless, radiologists should be aware of associated limitations including limited assessment of the inner left breast and pseudo-enhancement artifacts. KEY POINTS: • Breast MRI in patients with an implantable loop recorder is an infrequent, feasible, and safe procedure. • Despite limited breast visualization of the implantable loop recorder bed and characteristic artifacts, MRI depicted additional lesions in 8/40 (20%) of cases, half of which were malignant. • Breast MRI in patients with an implantable loop recorder should be performed when indicated, taking into consideration typical associated artifacts.

Breast MRI Background Parenchymal Enhancement Categorization Using Deep Learning: Outperforming the Radiologist.

BACKGROUND: Background parenchymal enhancement (BPE) is assessed on breast MRI reports as mandated by the Breast Imaging Reporting and Data System (BI-RADS) but is prone to inter and intrareader variation. Semiautomated and fully automated BPE assessment tools have been developed but none has surpassed radiologist BPE designations. PURPOSE: To develop a deep learning model for automated BPE classification and to compare its performance with current standard-of-care radiology report BPE designations. STUDY TYPE: Retrospective. POPULATION: Consecutive high-risk patients (i.e. >20% lifetime risk of breast cancer) who underwent contrast-enhanced screening breast MRI from October 2013 to January 2019. The study included 5224 breast MRIs, divided into 3998 training, 444 validation, and 782 testing exams. On radiology reports, 1286 exams were categorized as high BPE (i.e., marked or moderate) and 3938 as low BPE (i.e., mild or minimal). FIELD STRENGTH/SEQUENCE: A 1.5 T or 3 T system; one precontrast and three postcontrast phases of fat-saturated T1-weighted dynamic contrast-enhanced imaging. ASSESSMENT: Breast MRIs were used to develop two deep learning models (Slab artificial intelligence (AI); maximum intensity projection [MIP] AI) for BPE categorization using radiology report BPE labels. Models were tested on a heldout test sets using radiology report BPE and three-reader averaged consensus as the reference standards. STATISTICAL TESTS: Model performance was assessed using receiver operating characteristic curve analysis. Associations between high BPE and BI-RADS assessments were evaluated using McNemar's chi-square test (α* = 0.025). RESULTS: The Slab AI model significantly outperformed the MIP AI model across the full test set (area under the curve of 0.84 vs. 0.79) using the radiology report reference standard. Using three-reader consensus BPE labels reference standard, our AI model significantly outperformed radiology report BPE labels. Finally, the AI model was significantly more likely than the radiologist to assign "high BPE" to suspicious breast MRIs and significantly less likely than the radiologist to assign "high BPE" to negative breast MRIs. DATA CONCLUSION: Fully automated BPE assessments for breast MRIs could be more accurate than BPE assessments from radiology reports. LEVEL OF EVIDENCE: 4 TECHNICAL EFFICACY STAGE: 3.

Dipole modeling of multispectral signal for detecting metallic biopsy markers during MRI-guided breast biopsy: a pilot study.

PURPOSE: During MRI-guided breast biopsy, a metallic biopsy marker is deployed at the biopsy site to guide future interventions. Conventional MRI during biopsy cannot distinguish such markers from biopsy site air, and a post-biopsy mammogram is therefore performed to localize marker placement. The purpose of this pilot study is to develop dipole modeling of multispectral signal (DIMMS) as an MRI alternative to eliminate the cost, inefficiency, inconvenience, and ionizing radiation of a mammogram for biopsy marker localization. METHODS: DIMMS detects and localizes the biopsy marker by fitting the measured multispectral imaging (MSI) signal to the MRI signal model and marker properties. MSI was performed on phantoms containing titanium biopsy markers and air to illustrate the clinical challenge that DIMMS addresses and on 20 patients undergoing MRI-guided breast biopsy to assess DIMMS feasibility for marker detection. DIMMS was compared to conventional MSI field map thresholding, using the post-procedure mammogram as the reference standard. RESULTS: Biopsy markers were detected and localized in 20 of 20 cases using MSI with automated DIMMS post-processing (using a threshold of 0.7) and in 18 of 20 cases using MSI field mapping (using a threshold of 0.65 kHz). CONCLUSION: MSI with DIMMS post-processing is a feasible technique for biopsy marker detection and localization during MRI-guided breast biopsy. With a 2-min MSI scan, DIMMS is a promising MRI alternative to the standard-of-care post-biopsy mammogram.

Quantitative susceptibility mapping of the spine using in-phase echoes to initialize inhomogeneous field and R2* for the nonconvex optimization problem of fat-water separation.

Quantitative susceptibility mapping (QSM) of human spinal vertebrae from a multi-echo gradient-echo (GRE) sequence is challenging, because comparable amounts of fat and water in the vertebrae make it difficult to solve the nonconvex optimization problem of fat-water separation (R2*-IDEAL) for estimating the magnetic field induced by tissue susceptibility. We present an in-phase (IP) echo initialization of R2*-IDEAL for QSM in the spinal vertebrae. Ten healthy human subjects were recruited for spine MRI. A 3D multi-echo GRE sequence was implemented to acquire out-phase and IP echoes. For the IP method, the R2* and field maps estimated by separately fitting the magnitude and phase of IP echoes were used to initialize gradient search R2*-IDEAL to obtain final R2*, field, water, and fat maps, and the final field map was used to generate QSM. The IP method was compared with the existing Zero method (initializing the field to zero), VARPRO-GC (variable projection using graphcuts but still initializing the field to zero), and SPURS (simultaneous phase unwrapping and removal of chemical shift using graphcuts for initialization) on both simulation and in vivo data. The single peak fat model was also compared with the multi-peak fat model. There was no substantial difference on QSM between the single peak and multi-peak fat models, but there were marked differences among different initialization methods. The simulations demonstrated that IP provided the lowest error in the field map. Compared to Zero, VARPRO-GC and SPURS, the proposed IP method provided substantially improved spine QSM in all 10 subjects.

Free breathing three-dimensional cardiac quantitative susceptibility mapping for differential cardiac chamber blood oxygenation - initial validation in patients with cardiovascular disease inclusive of direct comparison to invasive catheterization.

BACKGROUND: Differential blood oxygenation between left (LV) and right ventricles (RV; ΔSaO2) is a key index of cardiac performance; LV dysfunction yields increased RV blood pool deoxygenation. Deoxyhemoglobin increases blood magnetic susceptibility, which can be measured using an emerging cardiovascular magnetic resonance (CMR) technique, Quantitative Susceptibility Mapping (QSM) - a concept previously demonstrated in healthy subjects using a breath-hold 2D imaging approach (2DBHQSM). This study tested utility of a novel 3D free-breathing QSM approach (3DNAVQSM) in normative controls, and validated 3DNAVQSM for non-invasive ΔSaO2 quantification in patients undergoing invasive cardiac catheterization (cath). METHODS: Initial control (n = 10) testing compared 2DBHQSM (ECG-triggered 2D gradient echo acquired at end-expiration) and 3DNAVQSM (ECG-triggered navigator gated gradient echo acquired in free breathing using a phase-ordered automatic window selection algorithm to partition data based on diaphragm position). Clinical testing was subsequently performed in patients being considered for cath, including 3DNAVQSM comparison to cine-CMR quantified LV function (n = 39), and invasive-cath quantified ΔSaO2 (n = 15). QSM was acquired using 3 T scanners; analysis was blinded to comparator tests (cine-CMR, cath). RESULTS: 3DNAVQSM generated interpretable QSM in all controls; 2DBHQSM was successful in 6/10. Among controls in whom both pulse sequences were successful, RV/LV susceptibility difference (and ΔSaO2) were not significantly different between 3DNAVQSM and 2DBHQSM (252 ± 39 ppb [17.5 ± 3.1%] vs. 211 ± 29 ppb [14.7 ± 2.0%]; p = 0.39). Acquisition times were 30% lower with 3DNAVQSM (4.7 ± 0.9 vs. 6.7 ± 0.5 min, p = 0.002), paralleling a trend towards lower LV mis-registration on 3DNAVQSM (p = 0.14). Among cardiac patients (63 ± 10y, 56% CAD) 3DNAVQSM was successful in 87% (34/39) and yielded higher ΔSaO2 (24.9 ± 6.1%) than in controls (p < 0.001). QSM-calculated ΔSaO2 was higher among patients with LV dysfunction as measured on cine-CMR based on left ventricular ejection fraction (29.4 ± 5.9% vs. 20.9 ± 5.7%, p < 0.001) or stroke volume (27.9 ± 7.5% vs. 22.4 ± 5.5%, p = 0.013). Cath measurements (n = 15) obtained within a mean interval of 4 ± 3 days from CMR demonstrated 3DNAVQSM to yield high correlation (r = 0.87, p < 0.001), small bias (- 0.1%), and good limits of agreement (±8.6%) with invasively measured ΔSaO2. CONCLUSION: 3DNAVQSM provides a novel means of assessing cardiac performance. Differential susceptibility between the LV and RV is increased in patients with cine-CMR evidence of LV systolic dysfunction; QSM-quantified ΔSaO2 yields high correlation and good agreement with the reference of invasively-quantified ΔSaO2.

The clinical utility of QSM: disease diagnosis, medical management, and surgical planning.

Quantitative susceptibility mapping (QSM) is an MR technique that depicts and quantifies magnetic susceptibility sources. Mapping iron, the dominant susceptibility source in the brain, has many important clinical applications. Herein, we review QSM applications in the diagnosis, medical management, and surgical treatment of disease. To assist in early disease diagnosis, QSM can identify elevated iron levels in the motor cortex of amyotrophic lateral sclerosis patients, in the substantia nigra of Parkinson's disease (PD) patients, in the globus pallidus, putamen, and caudate of Huntington's disease patients, and in the basal ganglia of Wilson's disease patients. Additionally, QSM can distinguish between hemorrhage and calcification, which could prove useful in tumor subclassification, and can measure microbleeds in traumatic brain injury patients. In guiding medical management, QSM can be used to monitor iron chelation therapy in PD patients, to monitor smoldering inflammation of multiple sclerosis (MS) lesions after the blood-brain barrier (BBB) seals, to monitor active inflammation of MS lesions before the BBB seals without using gadolinium, and to monitor hematoma volume in intracerebral hemorrhage. QSM can also guide neurosurgical treatment. Neurosurgeons require accurate depiction of the subthalamic nucleus, a tiny deep gray matter nucleus, prior to inserting deep brain stimulation electrodes into the brains of PD patients. QSM is arguably the best imaging tool for depiction of the subthalamic nucleus. Finally, we discuss future directions, including bone QSM, cardiac QSM, and using QSM to map cerebral metabolic rate of oxygen. Copyright © 2016 John Wiley & Sons, Ltd.

Reproducibility of quantitative susceptibility mapping in the brain at two field strengths from two vendors.

PURPOSE: To assess the reproducibility of brain quantitative susceptibility mapping (QSM) in healthy subjects and in patients with multiple sclerosis (MS) on 1.5 and 3T scanners from two vendors. MATERIALS AND METHODS: Ten healthy volunteers and 10 patients were scanned twice on a 3T scanner from one vendor. The healthy volunteers were also scanned on a 1.5T scanner from the same vendor and on a 3T scanner from a second vendor. Similar imaging parameters were used for all scans. QSM images were reconstructed using a recently developed nonlinear morphology-enabled dipole inversion (MEDI) algorithm with L1 regularization. Region-of-interest (ROI) measurements were obtained for 20 major brain structures. Reproducibility was evaluated with voxel-wise and ROI-based Bland-Altman plots and linear correlation analysis. RESULTS: ROI-based QSM measurements showed excellent correlation between all repeated scans (correlation coefficient R ≥ 0.97), with a mean difference of less than 1.24 ppb (healthy subjects) and 4.15 ppb (patients), and 95% limits of agreements of within -25.5 to 25.0 ppb (healthy subjects) and -35.8 to 27.6 ppb (patients). Voxel-based QSM measurements had a good correlation (0.64 ≤ R ≤ 0.88) and limits of agreements of -60 to 60 ppb or less. CONCLUSION: Brain QSM measurements have good interscanner and same-scanner reproducibility for healthy and MS subjects, respectively, on the systems evaluated in this study.

Multiple sclerosis lesion geometry in quantitative susceptibility mapping (QSM) and phase imaging.

PURPOSE: To demonstrate the phase and quantitative susceptibility mapping (QSM) patterns created by solid and shell spatial distributions of magnetic susceptibility in multiple sclerosis (MS) lesions. MATERIALS AND METHODS: Numerical simulations and experimental phantoms of solid- and shell-shaped magnetic susceptibility sources were used to generate magnitude, phase, and QSM images. Imaging of 20 consecutive MS patients was also reviewed for this Institutional Review Board (IRB)-approved MRI study to identify the appearance of solid and shell lesions on phase and QSM images. RESULTS: Solid and shell susceptibility sources were correctly reconstructed in QSM images, while the corresponding phase images depicted both geometries with shell-like patterns, making the underlying susceptibility distribution difficult to determine using phase alone. In MS patients, of the 60 largest lesions identified on T2 , 30 lesions were detected on both QSM and phase, of which 83% were solid and 17% were shells on QSM, and of which 30% were solid and 70% were shell on phase. Of the 21 shell-like lesions on phase, 76% appeared solid on QSM, 24% appeared shell on QSM. Of the five shell-like lesions on QSM, all were shell-like on phase. CONCLUSION: QSM accurately depicts both solid and shell patterns of magnetic susceptibility, while phase imaging fails to distinguish them.

Robustness of Deep Networks for Mammography: Replication Across Public Datasets.

Deep neural networks have demonstrated promising performance in screening mammography with recent studies reporting performance at or above the level of trained radiologists on internal datasets. However, it remains unclear whether the performance of these trained models is robust and replicates across external datasets. In this study, we evaluate four state-of-the-art publicly available models using four publicly available mammography datasets (CBIS-DDSM, INbreast, CMMD, OMI-DB). Where test data was available, published results were replicated. The best-performing model, which achieved an area under the ROC curve (AUC) of 0.88 on internal data from NYU, achieved here an AUC of 0.9 on the external CMMD dataset (N = 826 exams). On the larger OMI-DB dataset (N = 11,440 exams), it achieved an AUC of 0.84 but did not match the performance of individual radiologists (at a specificity of 0.92, the sensitivity was 0.97 for the radiologist and 0.53 for the network for a 1-year follow-up). The network showed higher performance for in situ cancers, as opposed to invasive cancers. Among invasive cancers, it was relatively weaker at identifying asymmetries and was relatively stronger at identifying masses. The three other trained models that we evaluated all performed poorly on external datasets. Independent validation of trained models is an essential step to ensure safe and reliable use. Future progress in AI for mammography may depend on a concerted effort to make larger datasets publicly available that span multiple clinical sites.

Artificial Intelligence-Enhanced Breast MRI: Applications in Breast Cancer Primary Treatment Response Assessment and Prediction.

ABSTRACT: Primary systemic therapy (PST) is the treatment of choice in patients with locally advanced breast cancer and is nowadays also often used in patients with early-stage breast cancer. Although imaging remains pivotal to assess response to PST accurately, the use of imaging to predict response to PST has the potential to not only better prognostication but also allow the de-escalation or omission of potentially toxic treatment with undesirable adverse effects, the accelerated implementation of new targeted therapies, and the mitigation of surgical delays in selected patients. In response to the limited ability of radiologists to predict response to PST via qualitative, subjective assessments of tumors on magnetic resonance imaging (MRI), artificial intelligence-enhanced MRI with classical machine learning, and in more recent times, deep learning, have been used with promising results to predict response, both before the start of PST and in the early stages of treatment. This review provides an overview of the current applications of artificial intelligence to MRI in assessing and predicting response to PST, and discusses the challenges and limitations of their clinical implementation.

[WITHDRAWN] Predicting breast cancer with AI for individual risk-adjusted MRI screening and early detection.

This paper has been withdrawn by Lukas Hirsch. Major revisions and rewriting in progress.

Improved subthalamic nucleus depiction with quantitative susceptibility mapping.

PURPOSE: To assess quantitative susceptibility mapping (QSM) in the depiction of the subthalamic nucleus (STN) by using 3-T magnetic resonance (MR) imaging. MATERIALS AND METHODS: This study was HIPAA compliant and institutional review board approved. Ten healthy subjects (five men, five women; mean age, 24 years ± 3 [standard deviation]; age range, 21-33 years) and eight patients with Parkinson disease (five men, three women; mean age, 57 years ± 14; age range, 25-69 years) who were referred by neurologists for preoperative navigation MR imaging prior to deep brain stimulator placement were included in this study. T2-weighted (T2w), T2*-weighted (T2*w), R2* mapping (R2*), phase, susceptibility-weighted (SW), and QSM images were reconstructed for STN depiction. Qualitative visualization scores of STN and internal globus pallidus (GPi) were recorded by two neuroradiologists on all images. Contrast-to-noise ratios (CNRs) of the STN and GPi were also measured. Measurement differences were assessed by using the Wilcoxon rank sum test and the signed rank test. RESULTS: Qualitative scores were significantly higher on QSM images than on T2w, T2*w, R2*, phase, or SW images (P < .05) for STN and GPi visualization. Median CNR was 6.4 and 10.7 times higher on QSM images than on T2w images for differentiation of STN from the zona incerta and substantia nigra, respectively, and was 22.7 and 9.1 times higher on QSM images than on T2w images for differentiation of GPi from the internal capsule and external globus pallidus, respectively. CNR differences between QSM images and all other images were significant (P < .01). CONCLUSION: QSM at 3-T MR imaging performs significantly better than current standard-of-care sequences in the depiction of the STN.

Automated Triage of Screening Breast MRI Examinations in High-Risk Women Using an Ensemble Deep Learning Model.

OBJECTIVES: The aim of the study is to develop and evaluate the performance of a deep learning (DL) model to triage breast magnetic resonance imaging (MRI) findings in high-risk patients without missing any cancers. MATERIALS AND METHODS: In this retrospective study, 16,535 consecutive contrast-enhanced MRIs performed in 8354 women from January 2013 to January 2019 were collected. From 3 New York imaging sites, 14,768 MRIs were used for the training and validation data set, and 80 randomly selected MRIs were used for a reader study test data set. From 3 New Jersey imaging sites, 1687 MRIs (1441 screening MRIs and 246 MRIs performed in recently diagnosed breast cancer patients) were used for an external validation data set. The DL model was trained to classify maximum intensity projection images as "extremely low suspicion" or "possibly suspicious." Deep learning model evaluation (workload reduction, sensitivity, specificity) was performed on the external validation data set, using a histopathology reference standard. A reader study was performed to compare DL model performance to fellowship-trained breast imaging radiologists. RESULTS: In the external validation data set, the DL model triaged 159/1441 of screening MRIs as "extremely low suspicion" without missing a single cancer, yielding a workload reduction of 11%, a specificity of 11.5%, and a sensitivity of 100%. The model correctly triaged 246/246 (100% sensitivity) of MRIs in recently diagnosed patients as "possibly suspicious." In the reader study, 2 readers classified MRIs with a specificity of 93.62% and 91.49%, respectively, and missed 0 and 1 cancer, respectively. On the other hand, the DL model classified MRIs with a specificity of 19.15% and missed 0 cancers, highlighting its potential use not as an independent reader but as a triage tool. CONCLUSIONS: Our automated DL model triages a subset of screening breast MRIs as "extremely low suspicion" without misclassifying any cancer cases. This tool may be used to reduce workload in standalone mode, to shunt low suspicion cases to designated radiologists or to the end of the workday, or to serve as base model for other downstream AI tools.

High-Temporal/High-Spatial Resolution Breast Magnetic Resonance Imaging Improves Diagnostic Accuracy Compared With Standard Breast Magnetic Resonance Imaging in Patients With High Background Parenchymal Enhancement.

PURPOSE: To compare breast magnetic resonance imaging (MRI) diagnostic performance using a standard high-spatial resolution protocol versus a simultaneous high-temporal/high-spatial resolution (HTHS) protocol in women with high levels of background parenchymal enhancement (BPE). MATERIALS AND METHODS: We conducted a retrospective study of contrast-enhanced breast MRIs performed at our institution before and after the introduction of the HTHS protocol. We compared diagnostic performance of the HTHS and standard protocol by comparing cancer detection rate (CDR) and positive predictive value of biopsy (PPV3) among women with high BPE (ie, marked or moderate). RESULTS: Among women with high BPE, the HTHS protocol demonstrated increased CDR (23.6 per 1,000 patients v 7.9 per 1,000 patients; P = 0. 013) and increased PPV3 (16.0% v 6.3%; P = .021) compared with the standard protocol. This corresponded to a 9.8% (95% CI, 1.29 to 18.3) decrease in the proportion of unnecessary biopsies among high-BPE patients and an additional cancer yield of 15.7 per 1,000 patients (95% CI, 1.3 to 18.3). CONCLUSION: Among women with high BPE, HTHS MRI improved diagnostic performance, leading to an additional cancer yield of 15.7 cancers per 1,000 women and concomitantly decreasing unnecessary biopsies by 9.8%. A multisite prospective trial is warranted to confirm these findings and to pave the way for more widespread clinical implementation.

Deep learning in breast imaging.

Millions of breast imaging exams are performed each year in an effort to reduce the morbidity and mortality of breast cancer. Breast imaging exams are performed for cancer screening, diagnostic work-up of suspicious findings, evaluating extent of disease in recently diagnosed breast cancer patients, and determining treatment response. Yet, the interpretation of breast imaging can be subjective, tedious, time-consuming, and prone to human error. Retrospective and small reader studies suggest that deep learning (DL) has great potential to perform medical imaging tasks at or above human-level performance, and may be used to automate aspects of the breast cancer screening process, improve cancer detection rates, decrease unnecessary callbacks and biopsies, optimize patient risk assessment, and open up new possibilities for disease prognostication. Prospective trials are urgently needed to validate these proposed tools, paving the way for real-world clinical use. New regulatory frameworks must also be developed to address the unique ethical, medicolegal, and quality control issues that DL algorithms present. In this article, we review the basics of DL, describe recent DL breast imaging applications including cancer detection and risk prediction, and discuss the challenges and future directions of artificial intelligence-based systems in the field of breast cancer.