| --- |
| tags: |
| - atrial-fibrillation |
| - ppg |
| - pytorch |
| --- |
| |
| ### Model Overview |
| This model is designed for **Atrial Fibrillation (AFib) detection** and **signal quality assessment** from Photoplethysmography (PPG) waveforms. |
|
|
| ### Input Specifications |
| * **Signal Type:** PPG |
| * **Sampling Rate:** 32 Hz |
| * **Input Shape:** `(batch_size, 1, 800)` |
|
|
| ### Required Preprocessing |
| To ensure optimal performance, input signals must be preprocessed using the following pipeline: |
| 1. **Bandpass Filter:** 0.3–7 Hz (2nd-order Butterworth, zero-phase). |
| 2. **Normalization:** Min-max normalization to scale the signal between [0, 1]. |
|
|
| ### Model Outputs |
| The model provides three primary outputs: |
|
|
| | Output | Description | |
| | :--- | :--- | |
| | `afib_prob` | **Raw Probabilities:** | |
| | `afib_pred` | **Classification:** Binary prediction (1: Atrial Fibrillation, 0: Normal Sinus Rhythm). **Thresholds have been calibrated based on quality assessment to optimize detection sensitivity and specificity.** | |
| | `qa_pred` | **Quality Assessment:** Signal reliability score (2: Good, 1: Acceptable, 0: Poor). | |
|
|
| --- |
| ## Details |
| Detailed training, architecture diagrams, and the preprocessing pipeline are available in the [repository](https://github.com/blublunalnal/afib_detection#project-1-revised-deepbeat) |
|
|
