Answers are grounded in this paper and cite their sources.
BackgroundConventional breast cancer screening is limited by cost, access, and patient discomfort. Whether a lightweight machine-learning model trained on routine laboratory indicators can support opportunistic screening remains unsettled.MethodsA retrospective cohort of 13,285 individuals (1,216 ICD-10 C50 cases, 12,069 controls) from a Chinese tertiary hospital during 2022 to 2024 was analysed with a leakage-controlled pipeline. Six classical learners and LightGBM were trained on twenty indicators, then assessed by discrimination, calibration, decision-curve analysis, and zero-shot UK Biobank validation.ResultsThe lightweight LightGBM achieved an internal AUC of 0.987 with sensitivity 0.835, specificity 0.983, Brier 0.026, and an external AUC of 0.83, training in 5.6 s within 1.65 MB.ConclusionThe model is a candidate low-cost triage tool, pending multicentre prospective validation and recalibration.
No takes yet. Share an insight, caveat, or question.
Dai et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: