DrKryptoMed/cancerpath-africa

Explainable four-agent pipeline for cervical cancer screening prioritisation and referral navigation --- title: CancerPath-Africa🏥 colorFrom: blue colorTo: red sdk: gradio sdk_version: 6.14.0 app_file: app.py pinned: true license: mit --- # 🏥 CancerPath-Africa **An Explainable Multi-Agent AI Pipeline for Cervical Cancer Screening Outreach Prioritisation and Referral Navigation in Kenya** *AI in Healthcare Bootcamp 2026 — Capstone Project* --- ## 📋 Table of Contents - Overview - Clinical Problem - System Architecture - Agent Descriptions - Explainability Framework - Datasets - Key Findings - Repository Structure - Installation - Usage - Technical Stack - Limitations - Citation --- ## Overview CancerPath-Africa is a modular multi-agent orchestrated AI pipeline designed to support community health workers in Kenya with cervical cancer screening outreach prioritisation and referral navigation. The system integrates classical machine learning, LLM-powered reasoning, geospatial facility mapping, and explainable AI into a single end-to-end pipeline that is deployable in low-resource Sub-Saharan African healthcare settings. --- ## Clinical Problem Cervical cancer is the most prevalent cancer among Kenyan women yet is almost entirely preventable with early screening. Three critical gaps drive late-stage presentation: | Gap | Statistic | |-----|-----------| | Screening non-uptake | 86% of Kenyan women never screened (DHS 2022) | | Referral non-completion | 72.8% face significant distance barriers | | Oncologist shortage | < 1 specialist per million people in SSA | CancerPath-Africa targets all three gaps through AI-assisted community health worker support. --- ## System Architecture ``` Patient / Health Worker Input ↓ ┌─────────────────────────────────┐ │ ORCHESTRATOR (orchestrator.py) │ │ Central pipeline controller │ └─────────────────────────────────┘ ↓ ┌─────────────────────────────────┐ │ AGENT 1 — Screening Predictor │ │ Logistic Regression + SHAP │ │ Threshold: 0.30 │ └─────────────────────────────────┘ ↓ Risk Level Out …