Medical AI Analysis System - Backend

A comprehensive medical analysis system with CLI and REST API interfaces, featuring anti-hallucination detection, provider review workflows, and AgentDB-powered pattern learning.

Features

🛡️ Anti-Hallucination System

• Confidence Scoring: Multi-factor confidence assessment for all analyses
• Citation Verification: Automatic verification of medical literature citations
• Knowledge Base Validation: Cross-reference with trusted medical sources
• Contradiction Detection: Identify contradictory statements and recommendations
• Provider Approval: Automatic routing to healthcare providers for low-confidence results

🏥 CLI Interface

# Analyze medical condition
medai analyze "diabetes" -s "increased thirst" "frequent urination" --age 45

# Interactive mode
medai analyze --interactive

# Verify analysis confidence
medai verify <analysisId> --detailed

# Provider review
medai provider review <analysisId> --decision approved --comments "Diagnosis confirmed"

# Notify provider
medai provider notify <analysisId> --urgent

# Configuration
medai config show
medai config set minimumConfidence 0.75

🌐 REST API

Endpoints

POST /api/analyze - Submit medical query

{
  "condition": "Type 2 Diabetes",
  "symptoms": ["increased thirst", "frequent urination"],
  "patientContext": {
    "age": 45,
    "gender": "male",
    "medicalHistory": ["hypertension"],
    "currentMedications": ["lisinopril"]
  },
  "options": {
    "requireHighConfidence": true,
    "includeEmergencyCheck": true
  }
}

GET /api/analysis/:id - Get analysis results

POST /api/provider/review - Submit provider review

{
  "analysisId": "123e4567-e89b-12d3-a456-426614174000",
  "decision": "approved",
  "comments": "Diagnosis confirmed through lab results"
}

POST /api/provider/notify - Notify provider

{
  "analysisId": "123e4567-e89b-12d3-a456-426614174000",
  "urgent": true
}

GET /api/metrics - Get learning metrics

📡 WebSocket Support

Real-time updates for analysis progress:

const ws = new WebSocket('ws://localhost:3000');

ws.on('message', (data) => {
  const message = JSON.parse(data);
  if (message.type === 'analysis_update') {
    console.log(`Progress: ${message.payload.progress}%`);
    console.log(`Status: ${message.payload.status}`);
  }
});

🧠 AgentDB Integration

• Pattern Recognition: Learn from successful analyses
• Skill Library: Build reusable diagnostic skills
• Reflexion Learning: Improve from provider feedback
• Embedding Search: Find similar historical cases

Architecture

src/
├── cli/                    # CLI interface
│   └── index.ts           # Command definitions
├── api/                    # REST API
│   └── index.ts           # Express server & WebSocket
├── services/              # Business logic
│   ├── medical-analysis.service.ts
│   ├── anti-hallucination.service.ts
│   ├── agentdb-learning.service.ts
│   └── provider.service.ts
├── middleware/            # Express middleware
│   ├── auth.middleware.ts
│   └── logging.middleware.ts
├── types/                 # TypeScript types
│   └── medical.types.ts
└── utils/                 # Utilities
    └── config.ts

Installation

# Install dependencies
npm install

# Build
npm run build

# Run API server
npm start

# Run CLI
npm run cli -- analyze "condition" -s "symptom1" "symptom2"

Configuration

Environment Variables

PORT=3000
AGENTDB_PATH=./data/medical-learning.db
MIN_CONFIDENCE=0.70
NODE_ENV=production
ALLOWED_ORIGINS=http://localhost:3000,https://app.example.com

Configuration File

{
  "antiHallucination": {
    "minimumConfidence": 0.70,
    "requireProviderReviewThreshold": 0.75,
    "autoApproveThreshold": 0.90,
    "enableContradictionDetection": true,
    "enableCitationVerification": true
  },
  "providers": {
    "notificationEnabled": true,
    "autoAssignProvider": true,
    "requiredForHighRisk": true
  },
  "api": {
    "port": 3000,
    "enableWebSocket": true,
    "rateLimitPerMinute": 100
  },
  "learning": {
    "enablePatternLearning": true,
    "agentDbPath": "./data/medical-learning.db",
    "minimumPatternFrequency": 3
  }
}

Anti-Hallucination Pipeline

1. Analysis Generation - AI generates medical analysis
2. Confidence Scoring - Multi-factor confidence calculation
   • Diagnosis confidence (0.3 weight)
   • Citation verification (0.25 weight)
   • Knowledge base validation (0.2 weight)
   • Contradiction detection (0.15 weight)
   • Provider alignment (0.1 weight)
3. Warning Generation - Identify potential issues
   • Low confidence
   • Hallucination detection
   • Contradictions
   • Emergency situations
4. Provider Review - Route to provider if needed
5. Learning Update - Update AgentDB with outcomes

Security

• Authentication: API key and Bearer token support
• Rate Limiting: 100 requests/minute per IP
• CORS: Configurable origin whitelist
• Helmet: Security headers
• Input Validation: Request validation
• Logging: Comprehensive request/response logging

Provider Workflow

1. Analysis Triggers Review - Low confidence or high-risk
2. Provider Notification - Email, SMS, or push notification
3. Provider Reviews - Approve, reject, or modify
4. Learning Update - System learns from provider decisions
5. Pattern Recognition - Apply learnings to future cases

Development

# Run in development mode
npm run dev

# Type checking
npm run typecheck

# Linting
npm run lint

# Tests
npm run test

Integration Examples

Express App Integration

import { MedicalAnalysisAPI } from './src/api';

const api = new MedicalAnalysisAPI(3000);
api.start();

CLI in Node.js

import { exec } from 'child_process';

exec('medai analyze "diabetes" -s "thirst" --output json', (error, stdout) => {
  const result = JSON.parse(stdout);
  console.log(result);
});

WebSocket Client

import WebSocket from 'ws';

const ws = new WebSocket('ws://localhost:3000');

ws.on('open', () => {
  ws.send(JSON.stringify({ type: 'subscribe', analysisId: '123' }));
});

ws.on('message', (data) => {
  const message = JSON.parse(data.toString());
  console.log('Update:', message);
});

License

MIT

Contributors

Built with the SPARC methodology and AgentDB integration.