MediScan AI

One-Stop Clinical AI Platform

Live Demo: frontend-lake-three-20.vercel.app

Screen for 7+ diseases, analyze lab reports, check drug interactions, and chat with an AI health assistant — all in one platform.

---

Screenshots

Landing Page

Health Screening — Multi-Disease AI

Lab Report Analyzer

Medicine Analyzer & Drug Interactions

AI Health Assistant — Symptom Checker, Chat & Note Summarizer

Login & Registration

---

How It Was Built

MediScan AI started as a diabetes risk predictor and evolved into a full clinical intelligence platform. Here's the technical journey:

Phase 1 — ML Foundation: Trained a stacking ensemble (Random Forest + XGBoost + LightGBM + Gradient Boosting) on 70,692 CDC Diabetes Health Indicators records. Used SMOTETomek for class imbalance, RobustScaler for preprocessing, and optimized the decision threshold for maximum F1 score. The model achieves 83% ROC-AUC with 88% recall.

Phase 2 — Full-Stack Platform: Built the backend with Django REST Framework (JWT auth, role-based access, patient CRUD, assessment workflow) and the frontend with React 18 + TypeScript + Tailwind CSS + Framer Motion for a modern clinical UI with dark mode.

Phase 3 — Multi-Disease Expansion: Integrated Google Gemini AI to power risk screening for 6 additional diseases (heart, stroke, kidney, liver, lung, thyroid) with disease-specific health indicator forms. Diabetes still uses the local ML model for zero-latency predictions.

Phase 4 — Lab Reports & Medicines: Added a lab report analyzer with medical reference ranges for 6 panels (CBC, lipid, metabolic, liver, kidney, thyroid) plus Gemini Vision for PDF/image OCR extraction. Integrated OpenFDA's free drug API for medicine search and built an AI-powered drug interaction checker.

Phase 5 — NLP Health Assistant: Created a symptom checker (describe symptoms in natural language, get urgency assessment and possible conditions), a health chatbot for general questions, and a clinical note summarizer that extracts diagnoses, medications, and follow-up actions.

Phase 6 — Reliability & Scale: Implemented multi-key API pools with round-robin load balancing and automatic failover for both Gemini and OpenFDA APIs. Failed keys enter cooldown and are retried automatically.

---

Features

Multi-Disease Health Screening

• AI-powered risk screening for 7 diseases: Diabetes, Heart Disease, Stroke, Kidney Disease, Liver Disease, Lung Disease, Thyroid
• Diabetes uses local ML ensemble (no API cost); others powered by Gemini AI
• Risk score, factors, recommendations, and detailed clinical analysis

Lab Report Analyzer

• Manual entry with 6 lab panels: CBC, Lipid, Metabolic, Liver, Kidney, Thyroid
• Upload PDF/image reports — Gemini Vision extracts values automatically
• Rule-based flagging with medical reference ranges (low/normal/high/critical)
• AI-powered clinical interpretation

Medicine Analyzer

• Drug search powered by OpenFDA API (free, no key needed)
• Detailed drug info: uses, dosage, side effects, contraindications
• Drug interaction checker for up to 10 medications
• AI-enhanced explanations via Gemini

AI Health Assistant

• Symptom Checker: NLP-powered symptom analysis with condition suggestions and urgency assessment
• Health Chat: Conversational AI for general health questions
• Note Summarizer: Paste clinical notes, get structured output with diagnoses, medications, and follow-up

Diabetes ML Assessment

• 3-step wizard with 21 CDC health indicators
• Stacking ensemble: Random Forest + XGBoost + LightGBM + Gradient Boosting
• SHAP-based risk factor explanation and ensemble breakdown
• PDF report download

Patient Management & Analytics

• Full CRUD with search, filters, and bulk CSV import
• Assessment history with trend charts
• Dashboard with risk distribution, monthly trends, age group analysis
• Real-time activity feed

Security & Admin

• JWT auth with auto token refresh and rotation
• Role-based access: Admin, Doctor, Nurse, Receptionist
• Admin panel with doctor/patient management and audit logs
• Admin access controlled by dev22ashish@gmail.com

---

ML Architecture

CDC Health Survey Data (70,692 records)
            |
    SMOTETomek Resampling
            |
    +---------------------------+
    |      Base Learners        |
    |  Random Forest | XGBoost  |
    |  LightGBM     | Grad.Boost|
    +---------------------------+
            |
    Logistic Regression (Meta)
            |
    Optimal Threshold: 0.3256
            |
      Risk Prediction

Metric	Score
ROC-AUC	0.8303
F1-Score	0.7751
Recall	0.8765
CV AUC	0.8711

---

Tech Stack

Layer	Technology	Purpose
Backend	Django 4.2 + DRF	REST API
Frontend	React 18 + TypeScript	UI framework
AI	Google Gemini 2.0 Flash	Disease screening, lab analysis, NLP
ML	scikit-learn + XGBoost + LightGBM	Diabetes ensemble model
Drug Data	OpenFDA API	Medicine search & interaction data
Database	PostgreSQL (Render)	Data persistence
Auth	SimpleJWT	Token-based authentication
Styling	Tailwind CSS + Framer Motion	UI + animations
Charts	Recharts	Data visualization
Reports	jsPDF	PDF generation
Hosting	Vercel + Render	Frontend + Backend

---

Quick Start

Backend

cd backend
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # Fill in credentials
python manage.py migrate
python manage.py runserver

Frontend

cd frontend
npm install --legacy-peer-deps
echo "VITE_API_URL=http://localhost:8000" > .env.local
npm run dev

---

Environment Variables

Backend (.env)

DEBUG=True
SECRET_KEY=your-secret-key-min-50-chars
DB_NAME=mediscan_db
DB_USER=postgres
DB_PASSWORD=your-db-password
DB_HOST=localhost
DB_PORT=5432
ALLOWED_HOSTS=localhost,127.0.0.1
CORS_ALLOWED_ORIGINS=http://localhost:3000,http://localhost:5173
ADMIN_SECRET_CODE=your-admin-secret-code
FRONTEND_URL=http://localhost:5173
GEMINI_API_KEY=your-gemini-api-key
# Multiple Gemini keys for load balancing:
# GEMINI_API_KEYS=key1,key2,key3

Frontend (.env.local)

VITE_API_URL=http://localhost:8000

---

API Endpoints

Authentication:
  POST /api/auth/login/                    JWT login
  POST /api/auth/register/                 Register user
  POST /api/auth/refresh/                  Refresh token

Patients:
  GET  /api/patients/                      List patients
  POST /api/patients/                      Create patient
  POST /api/patients/assessments/create/   Diabetes ML assessment

Screening:
  GET  /api/screening/diseases/            List available diseases
  POST /api/screening/create/              Run AI disease screening

Lab Reports:
  GET  /api/reports/panels/                List lab test panels
  POST /api/reports/analyze/               Analyze manual values
  POST /api/reports/upload/                Upload + AI extraction

Medicines:
  GET  /api/medicines/search/?q=name       Search drugs (OpenFDA)
  GET  /api/medicines/{drug_name}/         Drug details + AI
  POST /api/medicines/interactions/        Check interactions

AI Assistant:
  POST /api/ai/symptoms/                   Symptom analysis
  POST /api/ai/chat/                       Health chatbot
  POST /api/ai/summarize-notes/            Clinical note summary

Analytics:
  GET  /api/analytics/summary/             Dashboard stats
  GET  /api/analytics/risk-distribution/   Risk distribution
  GET  /api/analytics/trends/              Monthly trends

Admin:
  GET  /api/admin-panel/dashboard/         Admin stats
  GET  /api/admin-panel/doctors/           Doctor management
  GET  /api/docs/                          Swagger UI

---

Project Structure

mediscan-ai/
├── backend/
│   ├── config/          # Django settings, URLs
│   ├── users/           # Auth, profiles, password reset
│   ├── patients/        # Patient & assessment models
│   ├── screening/       # Multi-disease AI screening
│   ├── reports/         # Lab report analysis + reference ranges
│   ├── medicines/       # Drug search, interactions (OpenFDA)
│   ├── ai_engine/       # Gemini AI client, NLP views
│   ├── ml_engine/       # ML training, prediction, SHAP
│   ├── analytics/       # Dashboard data aggregation
│   └── admin_panel/     # Hospital admin management
│
└── frontend/
    └── src/
        ├── pages/       # All route pages (13 pages)
        ├── components/  # Reusable UI + shadcn/ui
        └── lib/         # API client, PDF generation

---

Author

Devashish

---

_{Built with Django, React, Gemini AI, scikit-learn, and OpenFDA}