ECG Feature Extraction and Classification App

This application provides a modular, clinical-grade pipeline for ECG signal processing, feature extraction, and heartbeat classification. It includes a Tkinter-based GUI for interactive visualization and classification of ECG segments, powered by a pre-trained machine learning model.

---

🧠 Overview

The app performs the following tasks:

• Loads ECG signals from CSV or WFDB formats
• Detects R-peaks and segments heartbeats
• Extracts clinically relevant features (RR interval, QRS duration, P-wave width, etc.)
• Classifies each heartbeat using a trained ExtraTreesClassifier model
• Displays results and confidence scores in a responsive GUI

---

🖥️ GUI Features

• Upload File: Load ECG data from CSV or WFDB .hea files
• Classify Heartbeat: Start classification of ECG segments
• Pause/Continue: Control the classification flow
• Segment Navigation: Browse previous/next heartbeat segments
• Live Plot: Visualize ECG waveform with Matplotlib
• Results Panel: View predicted class, confidence, and extracted features

The GUI is built using ttkbootstrap for a modern dark-themed interface and matplotlib for waveform visualization. Classification runs asynchronously to keep the UI responsive.

---

🧬 Feature Extraction

Each heartbeat segment is analyzed to extract:

• rr_interval: Time between successive R-peaks
• r_peak_amp: Amplitude of the R-peak
• qrs_duration: Duration of the QRS complex
• p_wave_width: Width of the P-wave
• min_voltage, max_voltage, mean_voltage, std_voltage: Signal statistics

These features are computed using slope analysis, moving averages, and statistical summaries. The ECGFeatureExtractor class handles this logic.

---

🤖 Classification

The classifier uses a pre-trained ExtraTreesClassifier (trained by me on mit-bih database) model with:

• Scaler: StandardScaler for feature normalization
• Encoder: LabelEncoder for label transformation
• Model: KNeighborsClassifier trained on balanced ECG feature data

Supported Labels

• N: Normal beat
• V: Ventricular
• S: Supraventricular
• F: Fusion
• Q: Unknown
• U: Unknown

The classifier returns both the predicted label and a confidence score (0–100%).

---

📦 Data Sources

Supported formats:

• CSV: Plain signal values, one column, no header
• WFDB: MIT-BIH .hea files (converted internally)

The ECGDataLoader handles both formats and automatically detects R-peaks using scipy.signal.find_peaks.

---

🧪 Model Evaluation

The classifier module includes:

• get_score(x, y): Returns accuracy score on test data
• display_confusion_matrix(x, y): Plots confusion matrix using matplotlib

These utilities help validate model performance and interpret classification results.

---

🧰 Developer Notes

• Modular architecture with separation of concerns
• Threaded data loading to prevent UI blocking
• Compatible with Windows and Linux
• Uses joblib for model persistence and pandas for feature handling
• UI state management includes pause/resume logic and segment navigation
• Includes requirements.txt for reproducible setup