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: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.