Armstrong66/ppqfl-breast-cancer-screening

Privacy-Preserving Quantum Federated Learning for mammographic breast cancer screening in African and MENA populations — HQCNN + simulated QFL with differential privacy, uncertainty quantification, and cross-population external validation. # Privacy-Preserving Quantum Federated Learning for Breast Cancer Screening ### African and MENA Population Contexts > Global Health Artificial Intelligence and Computing Laboratory - KCCR, KNUST, Ghana. --- ## Overview This project implements a hybrid quantum-classical neural network (HQCNN) within a simulated Quantum Federated Learning (QFL) framework for privacy-preserving breast cancer screening. Three virtual Ghanaian hospital clients (Accra, Kumasi, Tamale) train locally; only VQC parameters are aggregated — raw patient data never leaves any node. ## Key contributions - First QFL pipeline benchmarked on African mammography data - HQCNN: MobileNetV2 feature extractor + Variational Quantum Circuit (9–25 params) - Differential privacy simulation (σ_dp sweep) with privacy-utility trade-off analysis - Uncertainty quantification: MC-Dropout + quantum shot variance - Cross-population external validation: SA → MENA generalisation gap - Temperature scaling for VQC calibration (ECE correction) --- ## Repository Structure ``` ppqfl-breast-cancer-screening/ │ ├── _1_eda.py # Phase 1: Data audit & EDA (Mendeley + KAU-BCMD) ├── _2a_baseline.py # Phase 2a: MobileNetV2 classical baseline ├── _2b_feature_pca.py # Phase 2b: Feature extraction + PCA → quantum bridge ├── _3_5_vqc.py # Phases 3–5: VQC design, Regime A/B, sweep, noise ├── _6_7_uq.py # Phases 6–7: Uncertainty quantification + temperature scaling ├── _8_9_qfl.py # Phases 8–9: Simulated QFL + differential privacy ├── _10_11_external_val.py # Phases 10–11: KAU external validation + ablation table │ ├── cache_check.py # Pipeline cache guard (skip completed stages) ├── run_pipeline.sh # Full pipeline orchestrator (nohup / screen ready) ├── setup_local.sh # Local machine setup (path migration + deps) │ ├── requirements.txt # Pi …