Background Mammography is the current standard for breast cancer screening. This study aimed to develop an artificial intelligence (AI) algorithm for diagnosis of breast cancer in mammography, and explore whether it could benefit radiologists by improving accuracy of diagnosis. Methods In this retrospective study, an AI algorithm was developed and validated with 170 230 mammography examinations collected from five institutions in South Korea, the USA, and the UK, including 36 468 cancer positive confirmed by biopsy, 59 544 benign confirmed by biopsy (8827 mammograms) or follow-up imaging (50 717 mammograms), and 74 218 normal. For the multicentre, observer-blinded, reader study, 320 mammograms (160 cancer positive, 64 benign, 96 normal) were independently obtained from two institutions. 14 radiologists participated as readers and assessed each mammogram in terms of likelihood of malignancy (LOM), location of malignancy, and necessity to recall the patient, first without and then with assistance of the AI algorithm. The performance of AI and radiologists was evaluated in terms of LOM-based area under the receiver operating characteristic curve (AUROC) and recall-based sensitivity and specificity. Findings The AI standalone performance was AUROC 0•959 (95% CI 0•952-0•966) overall, and 0•970 (0•963-0•978) in the South Korea dataset, 0•953 (0•938-0•968) in the USA dataset, and 0•938 (0•918-0•958) in the UK dataset. In the reader study, the performance level of AI was 0•940 (0•915-0•965), significantly higher than that of the radiologists without AI assistance (0•810, 95% CI 0•770-0•850; p<0•0001). With the assistance of AI, radiologists' performance was improved to 0•881 (0•850-0•911; p<0•0001). AI was more sensitive to detect cancers with mass (53 [90%] vs 46 [78%] of 59 cancers detected; p=0•044) or distortion or asymmetry (18 [90%] vs ten [50%] of 20 cancers detected; p=0•023) than radiologists. AI was better in detection of T1 cancers (73 [91%] vs 59 [74%] of 80; p=0•0039) or node-negative cancers (104 [87%] vs 88 [74%] of 119; p=0•0025) than radiologists. Interpretation The AI algorithm developed with large-scale mammography data showed better diagnostic performance in breast cancer detection compared with radiologists. The significant improvement in radiologists' performance when aided by AI supports application of AI to mammograms as a diagnostic support tool.

Purpose To determine the relationship between tumor heterogeneity assessed by means of magnetic resonance (MR) imaging texture analysis and survival outcomes in patients with primary breast cancer. Materials and Methods Between January and August 2010, texture analysis of the entire primary breast tumor in 203 patients was performed with T2-weighted and contrast material-enhanced T1-weighted subtraction MR imaging for preoperative staging. Histogram-based uniformity and entropy were calculated. To dichotomize texture parameters for survival analysis, the 10-fold cross-validation method was used to determine cutoff points in the receiver operating characteristic curve analysis. The Cox proportional hazards model and Kaplan-Meier analysis were used to determine the association of texture parameters and morphologic or volumetric information obtained at MR imaging or clinical-pathologic variables with recurrence-free survival (RFS). Results There were 26 events, including 22 recurrences (10 local-regional and 12 distant) and four deaths, with a mean follow-up time of 56.2 months. In multivariate analysis, a higher N stage (RFS hazard ratio, 11.15 [N3 stage]; P = .002, Bonferroni-adjusted α = .0167), triple-negative subtype (RFS hazard ratio, 16.91; P < .001, Bonferroni-adjusted α = .0167), high risk of T1 entropy (less than the cutoff values [mean, 5.057; range, 5.022-5.167], RFS hazard ratio, 4.55; P = .018), and T2 entropy (equal to or higher than the cutoff values [mean, 6.013; range, 6.004-6.035], RFS hazard ratio = 9.84; P = .001) were associated with worse outcomes. Conclusion Patients with breast cancers that appeared more heterogeneous on T2-weighted images (higher entropy) and those that appeared less heterogeneous on contrast-enhanced T1-weighted subtraction images (lower entropy) exhibited poorer RFS. RSNA, 2016 Online supplemental material is available for this article.