AI-Medical-Assistant

⸻

EKG_AI_System

A local-first EKG analysis system combining:

• Machine learning (signal-based rhythm classification)
• On-demand screen capture
• AI-assisted clinical interpretation
• Clear confidence gating and safety constraints

This project is designed for edge / local execution, minimizing cloud dependency for core analysis and supporting real-world clinical workflows.

⚠️ This is a research / educational prototype.
It is not a medical device and must not be used for clinical decision-making without proper validation and regulatory approval.

---

Project Structure

EKG_AISystem/ │ ├── ekg_analyzer.py # Screen capture + AI chat interface ├── logs/ # Captured frames + AI responses ├── .venv/ # Analyzer virtual environment │ ├── ekg_ml/ │ ├── train_baseline.py │ ├── predict_from_signal.py │ ├── predict_from_image.py │ ├── preview_beats.py │ ├── extract_beats.py │ ├── models/ │ ├── data/ │ ├── .venv/ # ML virtual environment │ └── README.md

Important:
This project intentionally uses two virtual environments:

• One for the Analyzer UI + screen capture
• One for the ML pipeline

This avoids dependency conflicts (NumPy, pandas, wfdb, OpenCV).

---

Requirements

• Windows 10/11
• Python 3.11 (recommended)
• PowerShell
• Git
• Internet access for OpenAI API (Analyzer only)

---

Environment Setup (Step-by-Step)

1️⃣ Clone the Repository

git clone https://github.com/YOUR_USERNAME/EKG_AI_System.git
cd EKG_AI_System


⸻

Analyzer Environment (Screen Capture + AI)

2️⃣ Create and Activate Analyzer venv

py -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1

3️⃣ Install Analyzer Dependencies

python -m pip install --upgrade pip
python -m pip install numpy==1.26.4 scipy matplotlib scikit-learn joblib
python -m pip install pandas==2.2.3 wfdb==4.2.0
python -m pip install opencv-python==4.10.0.84 pillow mss openai

4️⃣ Verify Analyzer Environment

python -c "import cv2, wfdb, pandas, sklearn; print('OK')"


⸻

ML Environment (Training + Inference)

5️⃣ Create and Activate ML venv

cd ekg_ml
py -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1

6️⃣ Install ML Dependencies

python -m pip install --upgrade pip
python -m pip install numpy==1.26.4 scipy matplotlib scikit-learn joblib
python -m pip install pandas==2.2.3 wfdb==4.2.0
python -m pip install opencv-python==4.10.0.84 pillow

7️⃣ Verify ML Environment

python -c "import numpy, pandas, wfdb, cv2, sklearn; print('OK')"


⸻

Using Two PowerShell Terminals (Recommended)

Open two PowerShell terminals in VS Code:

Terminal A — Analyzer

cd EKG_AISystem
.\.venv\Scripts\Activate.ps1
python ekg_analyzer.py

This launches:
	•	The capture control window
	•	The chat UI
	•	Snapshot logging to logs/

⸻

Terminal B — ML Inference

cd EKG_AISystem\ekg_ml
.\.venv\Scripts\Activate.ps1

This terminal is used to:
	•	Train models
	•	Run predictions on saved images or signals

⸻

Training the Baseline Model

python train_baseline.py

Expected output:
	•	Class distribution warning (normal — dataset imbalance)
	•	Classification report
	•	Model saved to:

ekg_ml/models/baseline_rf.joblib


⸻

Predicting From a Signal (NPZ)

python predict_from_signal.py

Outputs:
	•	Top rhythm predictions with confidence

⸻

Predicting From a Screenshot Image

From the repo root (Analyzer env):

python .\ekg_ml\predict_from_image.py --image .\logs\frame_YYYYMMDD_HHMMSS.png

Important notes:
	•	The image must exist
	•	The model expects 252 features, not raw pixels
	•	The script internally converts image → signal proxy

⸻

Common Errors & Fixes

❌ ModuleNotFoundError: No module named 'cv2'

Cause: Wrong virtual environment active
Fix: Activate the correct .venv before running

⸻

❌ ValueError: X has 65536 features, expected 252

Cause: Raw image passed directly into ML model
Fix: Use predict_from_image.py, not the RF model directly

⸻

❌ wfdb / pandas StringArray TypeError

Cause: pandas version mismatch
Fix: Ensure:

numpy==1.26.4
pandas==2.2.3
wfdb==4.2.0


⸻

❌ Python was not found (Microsoft Store)

Cause: Windows App Execution Alias
Fix:
	•	Settings → Apps → Advanced App Settings
	•	Disable Python aliases

⸻

❌ KeyboardInterrupt when viewing plots

Cause: Matplotlib window closed manually
Fix: This is safe — no action needed

⸻

Design Philosophy
	•	Local ML first
	•	AI as decision-support, not authority
	•	Confidence-aware outputs
	•	Clinician-in-the-loop by default
	•	Clear separation of UI and ML concerns

⸻

Disclaimer

This project is intended for:
	•	Research
	•	Education
	•	Prototyping
	•	Workflow exploration

It is not FDA-approved and must not be used for patient care.

⸻

Future Directions
	•	Beat-level waveform reconstruction from images
	•	Multimodal fusion (EKG + vitals + context)
	•	Hardware-integrated assistants
	•	Robotics and bedside device workflows
	•	Wearable AI interfaces

⸻

## License

This project is **closed-source proprietary software**.

No use, redistribution, modification, or commercial deployment is permitted
without explicit written authorization from the author.

See the `LICENSE` file for full terms.

---