VitalBeat: Multi-Modal Arrhythmia Classification Engine

VitalBeat is a sophisticated biometric framework designed to tackle the challenge of automated cardiac arrhythmia detection. By integrating Electrocardiogram (ECG) and Photoplethysmogram (PPG) data, the system leverages the electrical and mechanical signatures of the heart to provide a dual-verification diagnostic model.

🚀 Key Features & Technical Innovation

• Multi-Sensor Data Fusion: VitalBeat addresses the "False Alarm" problem in clinical monitoring by cross-referencing ECG electrical peaks (R-peaks) with PPG pulse waves (Systolic peaks).
• Deep Signal Denoising: Employs a multi-stage filtering pipeline including a Butterworth Bandpass Filter to isolate cardiac frequencies ($0.5\text{Hz} - 40\text{Hz}$) and a Notch Filter to eliminate $50/60\text{Hz}$ powerline interference.
• Dynamic Peak Analysis: Utilizes adaptive thresholding for robust peak detection, even in the presence of baseline wander or motion artifacts commonly found in mobile PPG data.
• Multi-Class Arrhythmia Intelligence: The engine is trained to differentiate between Sinus Rhythm, Tachycardia (abnormally fast), Bradycardia (abnormally slow), and irregular rhythms like Atrial Fibrillation.

🛠️ Tech Stack & Implementation Details

• Language: Python 3.x.
• Signal Processing: SciPy.signal for digital filter design and convolution-based smoothing.
• Feature Engineering: Custom algorithms to calculate RR-Intervals (ECG) and PP-Intervals (PPG), alongside standard deviation of normal-to-normal intervals (SDNN).
• Machine Learning: Scikit-learn implementation of ensemble classifiers (e.g., Random Forest or SVM) optimized via GridSearch for maximum F1-score.

🧪 Detailed Methodology

1. Pre-Processing & Artifact Removal

Raw biometric data is often corrupted by "Baseline Wander" (breathing) and "Muscle Noise" (EMG interference). VitalBeat cleans these signals using:

• Zero-phase filtering to prevent phase distortion of the QRS complex.
• Moving Average Smoothing to sharpen PPG pulse contours for better peak identification.

2. Feature Extraction (The "Cardiac Signature")

From each heartbeat, the following features are distilled:

• Temporal Features: Heart Rate (BPM), Pulse Arrival Time (PAT), and Peak-to-Peak intervals.
• Morphological Features: Pulse width, slope ratio of the PPG dicrotic notch, and QRS duration from ECG.

3. Classification & Evaluation

The model is evaluated using a 70/30 train-test split with k-fold cross-validation. Performance is measured through:

• Confusion Matrices: To track specific misclassifications between similar arrhythmias.
• ROC-AUC Curves: To measure the model's ability to distinguish between healthy and pathological states.

📂 Project Structure

/BSP-main
├── BSP_FINAL.ipynb       # Python Notebook: Pre-processing, EDA, and ML Pipeline
├── BSP_PPT.pptx          # Technical Presentation: Visualizing results and logic
├── PPG_Dataset.7z        # Raw Biometric Repository (ECG & PPG time-series)
├── README.md             # Documentation (This file)
└── Research_Report.pdf   # Theoretical background and Arrhythmia Case Study