saraludjwera-star/projet-cardiology-IA

AI-powered clinical decision support system for heart failure in low-resource settings. It combines supervised learning models and ECG features to assist diagnosis, risk stratification, treatment recommendation, and patient follow-up. ⸻ # projet-cardiology-IA # Intelligent Diagnosis of Heart Failure Final project for the Building AI course ## Summary An AI-based tool to assist doctors in diagnosing, stratifying risk, recommending treatment, and following up heart failure patients in low-resource settings. ## Background Heart failure is a frequent and serious condition, often diagnosed too late in low-resource countries. Health professionals lack adapted decision-support tools, which leads to delayed care, avoidable complications, and high mortality. My motivation comes from my experience as a physician in Africa, where I saw these challenges firsthand. This project aims to: * Support early diagnosis and risk stratification * Detect silent high-risk profiles without obvious symptoms * Recommend initial treatment and personalized follow-up * Gradually integrate telemedicine for remote care ## How is it used? The user (doctor or health professional) inputs clinical, biological, and ECG data through a simple interface. The AI system provides: - a predicted diagnosis, - risk stratification, - prognosis estimation, - an initial treatment proposal (or referral recommendation), - a personalized follow-up plan. The tool is designed for health workers in rural or semi-urban areas without easy access to a specialist. It will work on a computer or tablet, even with limited internet connectivity. ## Data sources and AI methods Input data: - Clinical data (age, sex, history, symptoms) - Biological data (creatinine, BNP, etc.) - ECG features (QRS duration, T wave, heart rate, anomalies...) AI techniques: - **Model 1: MFNN (Multi-layer Feedforward Neural Network)** - Input: clinical, biological, and ECG data - Output: diagnosis, risk stratification, prognosis - Technique: supervised learning - **Model 2: Therapeutic supervised model** - Input: output of MFNN (dx, risk, prognosis) - Output: recommended treatment (e.g., medication, refer) and follow-up plan - **Unsupervised phase (future)** - Goal: i …