Hughes, Kevin S.

Abstract Background There is growing interest in artificial intelligence (AI) models for predicting future breast cancer. We performed a systematic review of studies of mammography-based AI models for future breast cancer risk prediction to summarize current evidence, identify knowledge gaps, and inform future research directions. Methods We searched 6 databases for studies from January 1, 2012, to February 28, 2025, that evaluated mammography-based AI models for future breast cancer risk prediction. We extracted study design, participants’ race and ethnicity, geographic origin, mammogram type, vendor, prediction time frame, breast cancer type predicted, external validation, and exclusion of cancers diagnosed on the index screening mammogram. Areas under the receiver operating curve (AUCs) were summarized overall and by study characteristics. Results A total of 41 studies met our inclusion criteria. All studies were retrospective, and most used 2D mammograms (n = 37 studies) acquired using Hologic equipment (n = 25) and performed in the United States (n = 17); White, non-Hispanic women were most represented. Nearly all (n = 40) studies assessed discrimination performance with a median AUC of 0.71 for no longer than 2-year risk prediction, 0.72 for 3-4 years, and 0.71 for 5 years or more prediction. Median AUC was 0.75 for studies including index cancers vs 0.68 when excluded. Six studies reported model calibration performance ranging from good to overestimation of risk. Conclusion Future studies should evaluate models using digital breast tomosynthesis, examine performance for aggressive or advanced breast cancer, include diverse populations, and evaluate both discrimination and model calibration. Prospective evaluations are needed to determine the clinical utility of mammography-based AI models for personalized risk-based breast cancer screening before implementation.

PURPOSE Literature on clinical note mining has highlighted the superiority of machine learning (ML) over hand-crafted rules. Nevertheless, most studies assume the availability of large training sets, which is rarely the case. For this reason, in the clinical setting, rules are still common. We suggest 2 methods to leverage the knowledge encoded in pre-existing rules to inform ML decisions and obtain high performance, even with scarce annotations. METHODS We collected 501 prostate pathology reports from 6 American hospitals. Reports were split into 2,711 core segments, annotated with 20 attributes describing the histology, grade, extension, and location of tumors. The data set was split by institutions to generate a cross-institutional evaluation setting. We assessed 4 systems, namely a rule-based approach, an ML model, and 2 hybrid systems integrating the previous methods: a Rule as Feature model and a Classifier Confidence model. Several ML algorithms were tested, including logistic regression (LR), support vector machine (SVM), and eXtreme gradient boosting (XGB). RESULTS When training on data from a single institution, LR lags behind the rules by 3.5% (F1 score: 92.2% v 95.7%). Hybrid models, instead, obtain competitive results, with Classifier Confidence outperforming the rules by +0.5% (96.2%). When a larger amount of data from multiple institutions is used, LR improves by +1.5% over the rules (97.2%), whereas hybrid systems obtain +2.2% for Rule as Feature (97.7%) and +2.6% for Classifier Confidence (98.3%). Replacing LR with SVM or XGB yielded similar performance gains. CONCLUSION We developed methods to use pre-existing handcrafted rules to inform ML algorithms. These hybrid systems obtain better performance than either rules or ML models alone, even when training data are limited.