Breast cancer remains one of the leading causes of mortality among women worldwide, particularly in regions with limited access to advanced diagnostic resources. This study presents a decentralized and lightweight artificial intelligence (AI) framework designed to assist in the early triage of breast cancer within low-resource healthcare environments. The proposed model employs an optimized deep learning architecture that operates efficiently on constrained devices while maintaining high diagnostic reliability. A federated learning strategy enables decentralized model training, ensuring data privacy and reducing the dependency on centralized computing infrastructures. To evaluate its clinical relevance, a Decision Curve Analysis (DCA) was integrated, offering a quantitative measure of net benefit across varying risk thresholds. Experimental results demonstrate notable performance, achieving an accuracy of 96.4%, sensitivity of 94.7%, specificity of 95.2%, F1-score of 95.0, and an area under the curve (AUC) of 0.98. The DCA further indicates superior clinical decision support compared to conventional centralized approaches. These outcomes confirm that decentralized, lightweight AI systems can deliver scalable, privacy-preserving, and ethically responsible solutions for breast cancer triage. The proposed framework not only addresses the computational and infrastructural barriers of low-resource settings but also promotes equitable access to AI-driven diagnostic technologies, bridging the gap between advanced machine intelligence and accessible public healthcare.

Keywords: Decentralized AI, Breast Cancer Triage, Federated Learning, Lightweight Deep Learning, Decision Curve Analysis, Low-Resource Healthcare, Ethical AI