Faculty Bibliography

PURPOSE: To compare a novel multicoil compressed sensing technique with flexible temporal resolution, golden-angle radial sparse parallel (GRASP), to conventional fat-suppressed spoiled three-dimensional (3D) gradient-echo (volumetric interpolated breath-hold examination, VIBE) MRI in evaluating the conspicuity of benign and malignant breast lesions. MATERIALS AND METHODS: Between March and August 2015, 121 women (24-84 years; mean, 49.7 years) with 180 biopsy-proven benign and malignant lesions were imaged consecutively at 3.0 Tesla in a dynamic contrast-enhanced (DCE) MRI exam using sagittal T1-weighted fat-suppressed 3D VIBE in this Health Insurance Portability and Accountability Act-compliant, retrospective study. Subjects underwent MRI-guided breast biopsy (mean, 13 days [1-95 days]) using GRASP DCE-MRI, a fat-suppressed radial "stack-of-stars" 3D FLASH sequence with golden-angle ordering. Three readers independently evaluated breast lesions on both sequences. Statistical analysis included mixed models with generalized estimating equations, kappa-weighted coefficients and Fisher's exact test. RESULTS: All lesions demonstrated good conspicuity on VIBE and GRASP sequences (4.28 +/- 0.81 versus 3.65 +/- 1.22), with no significant difference in lesion detection (P = 0.248). VIBE had slightly higher lesion conspicuity than GRASP for all lesions, with VIBE 12.6% (0.63/5.0) more conspicuous (P < 0.001). Masses and nonmass enhancement (NME) were more conspicuous on VIBE (P < 0.001), with a larger difference for NME (14.2% versus 9.4% more conspicuous). Malignant lesions were more conspicuous than benign lesions (P < 0.001) on both sequences. CONCLUSION: GRASP DCE-MRI, a multicoil compressed sensing technique with high spatial resolution and flexible temporal resolution, has near-comparable performance to conventional VIBE imaging for breast lesion evaluation. LEVEL OF EVIDENCE: 3 J. Magn. Reson. Imaging 2016.

We present a deep convolutional neural network for breast cancer screening exam classification, trained and evaluated on over 200,000 exams (over 1,000,000 images). Our network achieves an AUC of 0.895 in predicting the presence of cancer in the breast, when tested on the screening population. We attribute the high accuracy to a few technical advances. (i) Our network's novel two-stage architecture and training procedure, which allows us to use a high-capacity patch-level network to learn from pixel-level labels alongside a network learning from macroscopic breast-level labels. (ii) A custom ResNet-based network used as a building block of our model, whose balance of depth and width is optimized for high-resolution medical images. (iii) Pretraining the network on screening BI-RADS classification, a related task with more noisy labels. (iv) Combining multiple input views in an optimal way among a number of possible choices. To validate our model, we conducted a reader study with 14 readers, each reading 720 screening mammogram exams, and show that our model is as accurate as experienced radiologists when presented with the same data. We also show that a hybrid model, averaging the probability of malignancy predicted by a radiologist with a prediction of our neural network, is more accurate than either of the two separately. To further understand our results, we conduct a thorough analysis of our network's performance on different subpopulations of the screening population, the model's design, training procedure, errors, and properties of its internal representations. Our best models are publicly available at https://github.com/nyukat/breastcancerclassifier.

3D imaging enables accurate diagnosis by providing spatial information about organ anatomy. However, using 3D images to train AI models is computationally challenging because they consist of 10x or 100x more pixels than their 2D counterparts. To be trained with high-resolution 3D images, convolutional neural networks resort to downsampling them or projecting them to 2D. We propose an effective alternative, a neural network that enables efficient classification of full-resolution 3D medical images. Compared to off-the-shelf convolutional neural networks, our network, 3D Globally-Aware Multiple Instance Classifier (3D-GMIC), uses 77.98%-90.05% less GPU memory and 91.23%-96.02% less computation. While it is trained only with image-level labels, without segmentation labels, it explains its predictions by providing pixel-level saliency maps. On a dataset collected at NYU Langone Health, including 85,526 patients with full-field 2D mammography (FFDM), synthetic 2D mammography, and 3D mammography, 3D-GMIC achieves an AUC of 0.831 (95% CI: 0.769-0.887) in classifying breasts with malignant findings using 3D mammography. This is comparable to the performance of GMIC on FFDM (0.816, 95% CI: 0.737-0.878) and synthetic 2D (0.826, 95% CI: 0.754-0.884), which demonstrates that 3D-GMIC successfully classified large 3D images despite focusing computation on a smaller percentage of its input compared to GMIC. Therefore, 3D-GMIC identifies and utilizes extremely small regions of interest from 3D images consisting of hundreds of millions of pixels, dramatically reducing associated computational challenges. 3D-GMIC generalizes well to BCS-DBT, an external dataset from Duke University Hospital, achieving an AUC of 0.848 (95% CI: 0.798-0.896).

Breast MRI has high sensitivity and negative predictive value, making it well suited to problem solving when other imaging modalities or physical examinations yield results that are inconclusive for the presence of breast cancer. Indications for problem-solving MRI include equivocal or uncertain imaging findings at mammography and/or US; suspicious nipple discharge or skin changes suspected to represent an abnormality when conventional imaging results are negative for cancer; lesions categorized as Breast Imaging Reporting and Data System 4, which are not amenable to biopsy; and discordant radiologic-pathologic findings after biopsy. MRI should not precede or replace careful diagnostic workup with mammography and US and should not be used when a biopsy can be safely performed. The role of MRI in characterizing calcifications is controversial, and management of calcifications should depend on their mammographic appearance because ductal carcinoma in situ may not appear enhancing on MR images. In addition, ductal carcinoma in situ detected solely with MRI is not associated with a higher likelihood of an upgrade to invasive cancer compared with ductal carcinoma in situ detected with other modalities. MRI for triage of high-risk lesions is a subject of ongoing investigation, with a possible future role for MRI in decreasing excisional biopsies. The accuracy of MRI is likely to increase with the use of advanced techniques such as deep learning, which will likely expand the indications for problem-solving MRI. ^©RSNA, 2023 Quiz questions for this article are available in the supplemental material.

OBJECTIVE:To test the performance of a novel machine learning-based breast density tool. The tool utilizes a convolutional neural network to predict the BI-RADS based density assessment of a study. The clinical density assessments of 33,000 mammographic examinations (164,000 images) from one academic medical center (Site A) were used for training. MATERIALS AND METHODS/METHODS:This was an IRB approved HIPAA compliant study performed at two academic medical centers. The validation data set was composed of 500 studies from one site (Site A) and 700 from another (Site B). At Site A, each study was assessed by three breast radiologists and the majority (consensus) assessment was used as truth. At Site B, if the tool agreed with the clinical reading, then it was considered to have correctly predicted the clinical reading. In cases where the tool and the clinical reading disagreed, then the study was evaluated by three radiologists and the consensus reading was used as the clinical reading. RESULTS:For the classification into the four categories of the Breast Imaging Reporting and Data System (BI-RADS®), the AI classifier had an accuracy of 84.6% at Site A and 89.7% at Site B. For binary classification (dense vs. non-dense), the AI classifier had an accuracy of 94.4% at Site A and 97.4% at Site B. In no case did the classifier disagree with the consensus reading by more than one category. CONCLUSIONS:The automated breast density tool showed high agreement with radiologists' assessments of breast density.

Pathology reports are considered the gold standard in medical research due to their comprehensive and accurate diagnostic information. Natural language processing (NLP) techniques have been developed to automate information extraction from pathology reports. However, existing studies suffer from two significant limitations. First, they typically frame their tasks as report classification, which restricts the granularity of extracted information. Second, they often fail to generalize to unseen reports due to variations in language, negation, and human error. To overcome these challenges, we propose a BERT (bidirectional encoder representations from transformers) named entity recognition (NER) system to extract key diagnostic elements from pathology reports. We also introduce four data augmentation methods to improve the robustness of our model. Trained and evaluated on 1438 annotated breast pathology reports, acquired from a large medical center in the United States, our BERT model trained with data augmentation achieves an entity F1-score of 0.916 on an internal test set, surpassing the BERT baseline (0.843). We further assessed the model's generalizability using an external validation dataset from the United Arab Emirates, where our model maintained satisfactory performance (F1-score 0.860). Our findings demonstrate that our NER systems can effectively extract fine-grained information from widely diverse medical reports, offering the potential for large-scale information extraction in a wide range of medical and AI research. We publish our code at https://github.com/nyukat/pathology_extraction.

Breast cancer is the most common cancer in women, with the incidence rising substantially with age. Older women are a vulnerable population at increased risk of developing and dying from breast cancer. However, women aged 75 years and older were excluded from all randomized controlled screening trials, so the best available data regarding screening benefits and risks in this age group are from observational studies and modeling predictions. Benefits of screening in older women are the same as those in younger women: early detection of smaller lower-stage cancers, resulting in less invasive treatment and lower morbidity and mortality. Mammography performs significantly better in older women with higher sensitivity, specificity, cancer detection rate, and positive predictive values, accompanied by lower recall rates and false positives. The overdiagnosis rate is low, with benefits outweighing risks until age 90 years. Although there are conflicting national and international guidelines about whether to continue screening mammography in women beyond age 74 years, clinicians can use shared decision making to help women make decisions about screening and fully engage them in the screening process. For women aged 75 years and older in good health, continuing annual screening mammography will save the most lives. An informed discussion of the benefits and risks of screening mammography in older women needs to include each woman's individual values, overall health status, and comorbidities. This article will review the benefits, risks, and controversies surrounding screening mammography in women 75 years old and older and compare the current recommendations for screening this population from national and international professional organizations. ^©RSNA, 2023 Quiz questions for this article are available through the Online Learning Center.

The use of digital breast tomosynthesis (DBT) in breast cancer screening has become widely accepted, facilitating increased cancer detection and lower recall rates compared with those achieved by using full-field digital mammography (DM). However, the use of DBT, as compared with DM, raises new challenges, including a larger number of acquired images and thus longer interpretation times. While most current artificial intelligence (AI) applications are developed for DM, there are multiple potential opportunities for AI to augment the benefits of DBT. During the diagnostic steps of lesion detection, characterization, and classification, AI algorithms may not only assist in the detection of indeterminate or suspicious findings but also aid in predicting the likelihood of malignancy for a particular lesion. During image acquisition and processing, AI algorithms may help reduce radiation dose and improve lesion conspicuity on synthetic two-dimensional DM images. The use of AI algorithms may also improve workflow efficiency and decrease the radiologist's interpretation time. There has been significant growth in research that applies AI to DBT, with several algorithms approved by the U.S. Food and Drug Administration for clinical implementation. Further development of AI models for DBT has the potential to lead to improved practice efficiency and ultimately improved patient health outcomes of breast cancer screening and diagnostic evaluation. See the invited commentary by Bahl in this issue. ^©RSNA, 2022.

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has a high sensitivity in detecting breast cancer but often leads to unnecessary biopsies and patient workup. We used a deep learning (DL) system to improve the overall accuracy of breast cancer diagnosis and personalize management of patients undergoing DCE-MRI. On the internal test set (n = 3936 exams), our system achieved an area under the receiver operating characteristic curve (AUROC) of 0.92 (95% CI: 0.92 to 0.93). In a retrospective reader study, there was no statistically significant difference (P = 0.19) between five board-certified breast radiologists and the DL system (mean ΔAUROC, +0.04 in favor of the DL system). Radiologists' performance improved when their predictions were averaged with DL's predictions [mean ΔAUPRC (area under the precision-recall curve), +0.07]. We demonstrated the generalizability of the DL system using multiple datasets from Poland and the United States. An additional reader study on a Polish dataset showed that the DL system was as robust to distribution shift as radiologists. In subgroup analysis, we observed consistent results across different cancer subtypes and patient demographics. Using decision curve analysis, we showed that the DL system can reduce unnecessary biopsies in the range of clinically relevant risk thresholds. This would lead to avoiding biopsies yielding benign results in up to 20% of all patients with BI-RADS category 4 lesions. Last, we performed an error analysis, investigating situations where DL predictions were mostly incorrect. This exploratory work creates a foundation for deployment and prospective analysis of DL-based models for breast MRI.