mcs 0.0.8

Paper - Pytorch

Medical Diagnosis Swarm Architecture

MedicalSwarm is a sophisticated medical diagnosis and coding platform that leverages a coordinated swarm of specialized AI agents to deliver comprehensive medical assessments, accurate ICD-10 coding, and detailed clinical documentation. The system employs a hierarchical approach with specialized agents working in concert to analyze patient data, generate diagnoses, and ensure coding compliance.

Key Features

Multi-Agent Architecture

• Chief Medical Officer: Coordinates diagnosis workflow and synthesizes findings
• Virologist: Specializes in viral disease analysis and progression
• Internist: Provides comprehensive internal medicine evaluation
• Medical Coder: Ensures accurate ICD-10 coding and compliance
• Diagnostic Synthesizer: Creates final integrated assessments

Enterprise Integration

• RAG (Retrieval-Augmented Generation) API support
• Comprehensive logging and telemetry
• Scalable batch processing capabilities
• Configurable output formats and storage

Clinical Documentation

• Automated ICD-10 code assignment
• Hierarchical Condition Category (HCC) coding
• Evidence-based diagnostic rationale
• Detailed clinical progression timelines

---

Installation

pip install mcs

Onboarding

To get started you must first set some envs in your .env

WORKSPACE_DIR="agent_workspace"
OPENAI_API_KEY="your_key"
MASTER_KEY="328928402" # your master key for security

Usage

from mcs.main import MedicalCoderSwarm
import json

if __name__ == "__main__":
  # Example patient case
  patient_case = """
  Patient: 45-year-old White Male
  Location: New York, NY

  Lab Results:
  - egfr 
  - 59 ml / min / 1.73
  - non african-american
  
  """
  
  swarm = MedicalCoderSwarm(patient_id="Patient-001", max_loops=1, patient_documentation="")
  
  swarm.run(task=patient_case)
  
  print(json.dumps(swarm.to_dict()))

Example with HIPPA Grade Security

import json
from mcs.main import MedicalCoderSwarm

if __name__ == "__main__":
    # Extended Example Patient Case
    patient_case = """
    Patient Information:
    - Name: John Doe
    - Age: 45
    - Gender: Male
    - Ethnicity: White
    - Location: New York, NY
    - BMI: 28.5 (Overweight)
    - Occupation: Office Worker

    Presenting Complaints:
    - Persistent fatigue for 3 months
    - Swelling in lower extremities
    - Difficulty concentrating (brain fog)
    - Increased frequency of urination

    Medical History:
    - Hypertension (diagnosed 5 years ago, poorly controlled)
    - Type 2 Diabetes Mellitus (diagnosed 2 years ago, HbA1c: 8.2%)
    - Family history of chronic kidney disease (mother)

    Current Medications:
    - Lisinopril 20 mg daily
    - Metformin 1000 mg twice daily
    - Atorvastatin 10 mg daily

    Lab Results:
    - eGFR: 59 ml/min/1.73m² (Non-African American)
    - Serum Creatinine: 1.5 mg/dL
    - BUN: 22 mg/dL
    - Potassium: 4.8 mmol/L
    - HbA1c: 8.2%
    - Urinalysis: Microalbuminuria detected (300 mg/g creatinine)

    Vital Signs:
    - Blood Pressure: 145/90 mmHg
    - Heart Rate: 78 bpm
    - Respiratory Rate: 16 bpm
    - Temperature: 98.6°F
    - Oxygen Saturation: 98%

    Differential Diagnoses to Explore:
    1. Chronic Kidney Disease (CKD) Stage 3
    2. Diabetic Nephropathy
    3. Secondary Hypertension (due to CKD)
    4. Fatigue related to poorly controlled diabetes

    Specialist Consultations Needed:
    - Nephrologist
    - Endocrinologist
    - Dietitian for diabetic and CKD management

    Initial Management Recommendations:
    - Optimize blood pressure control (<130/80 mmHg target for CKD)
    - Glycemic control improvement (target HbA1c <7%)
    - Lifestyle modifications: low-sodium, renal-friendly diet
    - Referral to nephrologist for further evaluation
    """

    # Initialize the MedicalCoderSwarm with the detailed patient case
    swarm = MedicalCoderSwarm(
        patient_id="Patient-001",
        max_loops=1,
        # patient_documentation=patient_case,
        output_folder_path="reports",
        key_storage_path="example_key.key",
    )

    # Run the swarm on the patient case
    output = swarm.run(task=patient_case)

    # Print the system's state after processing
    print(json.dumps(swarm.to_dict(), indent=4))

Architecture Overview

flowchart TB
    CMO[Chief Medical Officer] --> V[Virologist]
    V --> I[Internist]
    I --> MC[Medical Coder]
    MC --> S[Synthesizer]
    

Features

• Specialized Agent Roles: Each agent has specific medical expertise and responsibilities
• Structured Diagnostic Flow: Organized pipeline from initial assessment to final synthesis
• ICD-10 Coding Integration: Comprehensive medical coding at each diagnostic stage
• Automated Report Generation: Standardized medical and coding reports
• Evidence-Based Decision Making: Multi-stage verification and synthesis process

Agent Responsibilities

mindmap
  root((Medical Swarm))
    Chief Medical Officer
      Initial Assessment
      Coordinate Specialists
      Treatment Plans
      Lab Range Analysis
    Virologist
      Viral Analysis
      Disease Progression
      Risk Assessment
    Internist
      System Review
      Vitals Analysis
      Comorbidity Evaluation
    Medical Coder
      ICD-10 Assignment
      Coding Compliance
      Documentation Review
    Synthesizer
      Integration
      Reconciliation
      Final Assessment

Diagnostic Flow Process

sequenceDiagram
    participant P as Patient Case
    participant CMO as Chief Medical Officer
    participant V as Virologist
    participant I as Internist
    participant MC as Medical Coder
    participant S as Synthesizer
    
    P->>CMO: Initial Data
    CMO->>V: Preliminary Assessment
    V->>I: Viral Analysis
    I->>MC: Comprehensive Review
    MC->>S: Coded Diagnosis
    S->>P: Final Report

API Usage

We have established an api in the /api folder. To run the api locally you must git clone, and then run:

cd api

chmod +x bootup.sh

./bootup.sh

API Testing

When you launch your api you can run the tests to see if it works ;)

cd api

python3 test.py

Docker Usage

To build and run the Docker container for the Medical Coder Swarm, follow these steps:

1. Build the Docker Image:

docker build -t mcs .

2. Run the Docker Container:

docker run --rm mcs

Docker Compose

docker-compose up

To stop the services, run:

docker-compose down

Full Diagram

graph TB
    subgraph External["External Systems"]
        API[("FastAPI Endpoints")]
        RAG["RAG System"]
        SEC["Security Layer"]
    end

    subgraph Core["Core MCS"]
        MS["MedicalCoderSwarm"]
        AG["AgentRearrange"]
        
        subgraph Agents["Agent Network"]
            CMO["Chief Medical Officer"]
            VIR["Virologist"]
            INT["Internist"]
            MC["Medical Coder"]
            SYN["Synthesizer"]
            SUM["Summarizer"]
        end
        
        subgraph Security["Security Components"]
            SDH["SecureDataHandler"]
            KRP["KeyRotationPolicy"]
        end
        
        subgraph IO["I/O Management"]
            LOG["Logger"]
            FS["File System"]
        end
    end

    %% Connections
    API --> MS
    MS --> AG
    AG --> Agents
    RAG --> MS
    MS --> SDH
    SDH --> KRP
    MS --> LOG
    MS --> FS
    
    %% Agent Flow
    CMO --> VIR
    VIR --> INT
    INT --> MC
    MC --> SYN
    SYN --> SUM
    
    class API,RAG,SEC external
    class MS,AG core
    class SDH,KRP security
    class LOG,FS io

Lab Range Analysis

The system includes specialized functionality for analyzing lab results against diagnostic criteria:

• Automated range checking for common tests (e.g., eGFR)
• Diagnosis-specific range validation
• Multi-factor analysis for complex diagnoses

Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

• Built with the Swarms framework
• Utilizes GPT-4 for advanced medical reasoning
• ICD-10 coding standards compliance

Contact

For questions and support, please open an issue in the repository.