Documentation

Installation

1. Clone Repository

# Clone repository
git clone https://github.com/aim-lab/AFtoolkit.git
cd AFtoolkit

2. Create Conda Environment

# Create conda environment
conda create -n aftoolkit python=3.10 -y
conda activate aftoolkit

3. Install Dependencies

# Install dependencies
pip install poetry
poetry install

4. Verify Installation

# Verify installation
python -c "import tensorflow as tf; print(tf.__version__)"

Quick Start

import wfdb
import numpy as np

# Load ECG data
record = wfdb.rdrecord('04015', pn_dir='afdb/1.0.0')
beats = wfdb.rdann('04015', 'qrs', pn_dir='afdb/1.0.0')

# Extract RR intervals
rr_intervals = np.diff(beats.sample) / record.fs

# Run AF detection (example - adjust to actual API)
# results = arnet2.detect(rr_intervals)

Supported Input Formats

Required Fields

• RR intervals (in seconds or samples)
• Sampling rate (Hz)
• Recording duration

Output Format

{
  "af_burden_percent": 9.8,
  "episodes": [
    {
      "start": "02:14:32",
      "duration": "28m 15s",
      "confidence": 0.985
    },
    ...
  ],
  "phenotype": "Type I: Nocturnal" // if applicable
}

1. MIT-BIH Atrial Fibrillation Database

URL: physionet.org/content/afdb/1.0.0/

• 25 long-term recordings (10 hours each)
• Annotated AF episodes
• Free download

2. CPSC 2021

URL: physionet.org/content/cpsc2021/1.0.0/

• 730+ recordings
• Paroxysmal AF focus
• PhysioNet access

Download Instructions

Using wget:

wget -r -N -c -np https://physionet.org/files/afdb/1.0.0/

Using Python:

import wfdb
record = wfdb.rdrecord('04015', pn_dir='afdb/1.0.0')

Running AF Detection

# Example: Running ArNet2 on test data
from AFtoolkit.detection import run_ArNet2
import wfdb

# Load sample ECG data
record = wfdb.rdrecord('path/to/record')
annotations = wfdb.rdann('path/to/record', 'qrs')

# Run detection
results = run_ArNet2.detect(
    rr_intervals=annotations.sample,
    sampling_rate=record.fs
)

# Print results
print(f"AF Burden: {results['af_burden_percent']}%")
print(f"Episodes detected: {len(results['episodes'])}")

Running Phenotype Clustering

# Example: Phenotype prediction
from AFtoolkit.phenotyping import run_prediction

# Assuming you have 24-hour burden profile
hourly_burden = [...] # 24 values

# Predict phenotype
phenotype_result = run_prediction.predict(
    burden_profile=hourly_burden
)

print(f"Predicted phenotype: {phenotype_result['phenotype']}")
print(f"Cluster assignment: {phenotype_result['cluster']}")