Breast cancer is one of the most prevalent cancers affecting women globally. Early diagnosis is crucial for effective treatment and improved survival rates. Imaging techniques such as mammography and ultrasound are widely used conventional diagnostic methods. However, these methods have limitations, including low sensitivity and specificity, especially in patients with dense breast tissue. For instance, mammograms miss approximately 20% of breast cancer cases, leading to false negatives and delayed treatment that can have fatal consequences. To address these challenges, artificial intelligence (AI)-based diagnostic tools have been developed to assist healthcare professionals in accurately detecting breast cancer. These tools work in conjunction with human radiologists to improve diagnostic outcomes. In addition, biomarkers present a promising non-invasive, more convenient alternative for the early detection of breast cancer, potentially overcoming the limitations of traditional screening methods. Various biomarkers, such as circulating tumor cells, cell-free tumor nucleic acids, and microRNAs, have shown promise in early breast cancer diagnosis. A systematic literature review is needed to consolidate ongoing efforts in molecular biology and biomedical sciences aimed at achieving early breast cancer diagnosis. One of the limitations of previously published research is the heterogeneity of methodologies, which can compromise the credibility of comparisons due to potential inaccuracies in the original data. Hence, future studies should prioritize using consistent datasets and developing robust techniques to manage missing values, outliers, and class imbalances to improve the reliability of breast cancer detection models. This literature review seeks to bridge the knowledge gap by reporting recent high-performing AI models and effective biomarkers that can serve as diagnostic tools in clinical practice.

Choosing which features go into a clinical risk model is not only a statistical problem. A model that is accurate but whose feature list makes no clinical sense will not be trusted by the physicians it is meant to help, and most feature-selection methods still include a clinician’s judgment only as a filter applied after the search, not as part of it. This paper introduces KGEN, a genetic algorithm that builds an oncologist’s feature ratings directly into both the starting population and the fitness function, alongside data-driven importance scores, through one tunable weight. We tested this on 67,858 women from the PLCO Cancer Screening Trial, searching over 75 candidate features, against a matched no-expert control run under an identical search budget. Across four independent runs, expert guidance raised clinical relevance substantially and consistently (0.929–1.000 versus 0.682 for the no-expert control and 0.669–0.720 for four of the five conventional baselines, the exception being BCRAT, whose fixed feature set happens to reach a perfect 1.000; pooled $p=0.0019$ across sub-populations), while discrimination stayed within the same narrow range as every conventional baseline throughout, and, in three of the four sub-populations, was even modestly but significantly higher than the no-expert control’s: a clinically sensible feature list did not come at the cost of accuracy. The same wide-margin clinical-relevance pattern held on an independent hospital-readmission dataset of 99,343 encounters and an independent breast-cancer gene-expression cohort, each with a small, measurable accuracy trade-off against the strongest conventional baselines. Feature-selection stability across runs was also higher under expert guidance for three of four sub-populations. This suggests that a clinician’s judgment can be built into a search algorithm as a working input, not just a check performed afterward.

Background and Objectives: Accurate breast cancer risk prediction requires balancing predictive performance with clinical interpretability — a trade-off that existing methods address through post-hoc adjustment rather than principled optimisation. Widely used tools such as BCRAT rely onpredefined variables, yielding modest discrimination (AUC 0.55–0.65). Modern machine learning models achieve higher performance (AUC > 0.85) but select clinically implausible features, limiting physician adoption. No existing method formally incorporates domain expertise as a quantitative operator input to evolutionary feature selection. This study introduces an 𝛼-weighted expert-data fusion operator that encodes oncologist-derived feature relevance directly into genetic algorithm initialisation and fitness evaluation.Methods: An 𝛼-weighted fusion operator combines oncologist-derived feature relevance ratings with Random Forest importance scores to guide both population initialisation and fitness evaluation within a genetic algorithm (GA). Three GA variants were developed to explore the performance interpretability–parsimony trade-off: a Clinical GA with static expert weighting (𝛼 = 0.75), an Adaptive GA with dynamic objective scheduling, and a Multi-Population GA with specialised sub-populations and elite migration. The framework was evaluated on the PLCO Cancer Screening Trial cohort (𝑛 = 72,191) against five baselines using AUC, DeLong testing, and calibration metrics (Brier score, ECE, Hosmer–Lemeshow test). Results: The proposed framework achieved statistically significant improvements over all baselines (𝑝 < 0.001), with ΔAUC up to +0.447 vs. BCRAT and +0.231 vs. RFECV. Post-calibration performance remained stable (Brier score 0.018–0.024). The Clinical GA selected 79.3% clinically aligned features, while the Adaptive GA achieved peak discrimination (AUC 0.948). External validation on the METABRIC dataset showed a 26.2% improvement in clinical relevance, supporting transferability across datasets.Conclusion: Integrating quantitative expert knowledge into evolutionary optimisation enables systematic navigation of the performance–interpretability trade-off in clinical risk modelling. This framework provides a generalisable strategy for incorporating domain expertise into machine learning pipelines for clinical decision support.