Deep Neural Networks Improve Radiologists' Performance in Breast Cancer Screening.

We present a deep convolutional neural network for breast cancer screening exam classification, trained, and evaluated on over 200000 exams (over 1000000 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. 1) 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. 2) 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. 3) Pretraining the network on screening BI-RADS classification, a related task with more noisy labels. 4) 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/breast_cancer_classifier.

Overstated Harms of Breast Cancer Screening? A Large Outcomes Analysis of Complications Associated With 9-Gauge Stereotactic Vacuum-Assisted Breast Biopsy.

OBJECTIVE: The purpose of this study was to assess the rate, type, and severity of complications related to 9-gauge stereotactic vacuum-assisted breast biopsy (SVAB) and to delineate associated factors that may contribute to a higher rate of complications. MATERIALS AND METHODS: This retrospective study included 4776 patients who underwent SVAB between 2003 and 2016. A total of 319 patients with documented postbiopsy complications were identified. Complications were subcategorized as bleeding, pain, lightheadedness, bruising, and other complications, and their severity was classified as minor, moderate, or severe. Hematoma volumes were correlated with biopsy location and complication severity. A group of control subjects who underwent SVAB but had no complications was compared with the group of study patients with regard to age, biopsy location, lesion type, and pathologic findings. Postbiopsy screening adherence was assessed. Statistical analyses were performed using the Fisher exact, Mann-Whitney, Kruskal-Wallis, and Spearman rank correlation tests. RESULTS: Of the 319 patients with complications who were identified (representing 6.7% of the 4776 patients who underwent SVAB), 307 (96.2%) had mild complications, 12 (3.8%) had moderate complications, and no patients had severe complications. The most common complication was bleeding or hematoma (89.3% of patients [285/319]), followed by pain (6.9% [22/319]), lightheadedness (0.9% [3/319]), bruising (0.9% [3/319]), and other complications (1.9% [6/319]). No significant differences were noted between the study group and the control group in terms of age (p = 0.474), biopsy location (p = 0.065), histologic findings (p = 0.056), or lesion type (p = 0.568). Hematoma volume (median, 7.5 cm3) did not correspond to the severity of complications. Larger hematoma volumes were associated with a posterior biopsy location (p = 0.008). The rate of return to annual screening after biopsy was not adversely affected by the presence of biopsy complications. CONCLUSION: Clinically significant complications associated with SVAB were exceedingly rare (0.3%) in this large study spanning 13 years.