Artificial Intelligence Pipeline for Mammography-Based Breast Cancer Detection: An Integrated Systematic Review and...

Area of Science:

• Medical Imaging and Artificial Intelligence
• Oncology and Machine Learning
• Radiology and Computational Pathology

Background:

• Breast cancer detection using mammography is critical globally.
• Artificial intelligence (AI) offers potential for improving mammography interpretation.
• Existing AI pipelines lack interpretability and standardization.

Purpose of the Study:

• To systematically review current AI applications in mammography-based breast cancer detection.
• To experimentally validate representative AI models (CNNs, SVM, XGBoost) on a public dataset.
• To assess the interpretability and reproducibility of AI pipelines.

Main Methods:

• A PRISMA 2020-compliant systematic review of PubMed and Scopus/ScienceDirect.
• Experimental validation of Convolutional Neural Networks (CNNs) and classical machine learning models (SVM, XGBoost) on the CBIS-DDSM dataset.
• Application of explainable AI (XAI) methods, including Grad-CAM and SHAP, for interpretability within a machine learning operations (MLOps) framework.

Main Results:

• Systematic review identified heterogeneity in datasets, preprocessing, and validation, with limited use of XAI.
• ResNet-50 achieved the highest performance (AUC-ROC 0.95, sensitivity 89%), followed by XGBoost (AUC-ROC 0.90) and SVM (AUC-ROC 0.84).
• Grad-CAM provided plausible lesion-focused heatmaps; SHAP revealed simple image features drove classical model predictions.

Conclusions:

• Current AI pipelines show potential for mammography-based breast cancer detection but have limitations.
• Standardized preprocessing, rigorous external validation, and routine XAI are essential for clinical deployment.
• This study highlights the need for robust and interpretable AI in breast cancer diagnostics.