Faculty Bibliography

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.

Background Although previous studies have focused on rural disparities in the use of screening mammography, city-level use throughout the United States has not been well evaluated even though more than 30 million women live in the 500 largest cities. Purpose To evaluate disparities in the city-level use of screening mammography and to identify factors that have an impact on screening utilization. Materials and Methods This retrospective study used data from large publicly available databases, the American Community Survey and Robert Wood Johnson Foundation 500 Cities Project, which includes screening mammography utilization data from the Behavioral Risk Factor Surveillance System. Databases were searched from January to March 2018. The use of screening mammography was evaluated at the city level by census region and division by using the Mann-Whitney U test. Univariable Spearman rank correlation and multivariable regression analysis were performed to determine the impact of factors on screening use, including population size, health-related variables (use of Papanicolaou test, obesity), income variables (median household income, poverty status, health insurance), and race. Results Overall mean city-level screening mammography use rate was 77.7% (range, 62.8%-88.9%). The highest mean utilization occurred in coastal cities, with the highest overall utilization in the New England area (82.7%). The lowest utilization rate was in Mountain states (73.6%). City-level utilization showed a positive correlation with Papanicolaou test use (r = 0.75, P < .001), median household income (r = 0.44, P < .001), and percentage Asian population (r = 0.38, P < .001) and a negative correlation with obesity (r = -0.36, P < .001), the lack of health insurance (r = -0.44, P < .001), and poverty (r = -0.30, P < .001). Multivariable analysis showed the strongest independent predictors of utilization to be percentage of women screened with the Papanicolaou test, Asian race, private insurance, and census division (R² = 68%). Conclusion Disparities in the utilization of preventive health care services exist at the large city level, with the highest use in New England cities and lowest in Mountain cities. Predictors of higher than average utilization include census division and percentage of inhabitants who are up to date with the Papanicolaou test, are of Asian race, and have private insurance. © RSNA, 2019.

BACKGROUND:Dynamic contrast enhanced (DCE) breast MRI is highly sensitive for breast cancer and requires gadolinium-based contrast agents (GBCA)s, which have potential safety concerns. PURPOSE/OBJECTIVE:Test whether breast cancers imaged by 3T DCE breast MRI with 0.05 mmol/kg of gadobutrol are detectable. METHODS:Analysis of 3T DCE breast MRIs with half dose of gadobutrol from patients included in an IRB-approved and HIPPA-compliant prospective study of breast PET/MRI. Between 11/7/2014 and 3/2/2018, 41 consecutive women with biopsy-proven breast cancer that was at least 2 cm, multi-focal or multi-centric, had axillary metastasis, or had skin involvement who gave informed consent were included. Two breast radiologists independently recorded lesion conspicuity on a 4-point scale (0 = not seen, 1 = questionably seen, 2 = adequately seen, 3 = certainly seen), and measured the lesion. Size was compared between radiologists and with size on available mammogram, ultrasound, MRI, and surgical pathology. Inter-reader agreement was assessed by kappa coefficient for conspicuity. Lesion size comparisons were assessed using the Spearman rank correlation. RESULTS:In 40 patients (ages 28.4-80.5, 51.9 years), there were 49 cancers. 10.1% of lesions were 1 cm or less and 26.5% of lesions were 2 cm or less. Each reader detected 49/49 cancers. Conspicuity scores ranged from 2 to 3, mean 2.9/3 for both readers (p = 0.47). Size on half-dose 3T DCE-MRI correlated with size on surgical pathology (r = 0.6, p = 0.03) while size on mammogram and ultrasound did not (r = 0.25, p = 0.46; r = 0.25, p = 0.42). CONCLUSION/CONCLUSIONS:All breast cancers in this cohort, as small as 0.4 cm, were seen on 3T DCE breast MRI with 0.05 mmol/kg dose of gadobutrol.

Positron emission tomography / magnetic resonance imaging (PET/MRI) is an emerging imaging technology that allows for the acquisition of multiple MRI parameters simultaneously with PET data. In this review, we address the technical requirements of PET/MRI including protocols and tracers, the potential of integrated localized breast PET/MRI exams, and possible applications of whole-body PET/MRI in breast cancer patients. Currently, PET/MRI can be performed on sequential and integrated PET/MRI scanners but, as not all practices can access these dedicated machines, several studies look at PET and MRI exams that are performed separately on separate scanners within a short time frame. This practice likely provides similar clinical data, although exact colocalization for iso-voxel analysis, currently performed only in research, is not possible. In PET/MRI, the MRI sequences are flexible and can be customized according to the aim of the exam. The most commonly used radiotracer is ¹⁸ F-FDG; however, tracers that image hypoxia and drug targets such as estrogen receptors and HER2 are in development and may increase the utility of PET/MRI. For dedicated breast PET/MRI, a potential advantage over standard breast MRI alone may be the complementary sensitivities of MRI for extent of disease within the breast and PET for axillary and internal mammary nodal metastases. Moreover, layers of multiparametric MRI and PET metrics derived from the index lesion are being investigated as predictors of response to neoadjuvant therapy. These data may eventually be able to be quantified and mined in a way that furthers radiomics and also precision medicine. Finally, in whole-body imaging of breast cancer patients, single-institution studies have found that PET/MRI detects more metastases than PET at about half the radiation dose, although a survival benefit has not been shown. For now, whole-body PET/MRI in breast cancer patients may be most relevant for young patients who may undergo serial surveillance exams.