Upload 4 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+ecg_cnn_model.keras filter=lfs diff=lfs merge=lfs -text

Readme.md

@@ -0,0 +1,52 @@
+# CNN-Based Cardiac Abnormality Detection Using PTB-XL
+
+This repository contains a deep learning model for automatic detection of
+cardiac abnormalities from 12-lead ECG signals using the PTB-XL dataset.
+
+The model is trained using a 1D Convolutional Neural Network (CNN) and
+supports multi-label classification.
+
+---
+
+## 🔍 Supported Diagnostic Classes
+
+- **NORM** – Normal ECG  
+- **MI** – Myocardial Infarction  
+- **STTC** – ST/T Wave Changes  
+- **CD** – Conduction Disturbance  
+- **HYP** – Hypertrophy  
+
+---
+
+## 📥 Input Format
+
+- Shape: **(1000, 12)**
+- Sampling rate: **100 Hz**
+- Duration: **10 seconds**
+- Preprocessing:
+  - Bandpass filtering (0.5–40 Hz)
+  - Z-score normalization (per lead)
+
+---
+
+## 🧠 Model Architecture
+
+- 1D CNN with multiple convolutional blocks
+- Batch normalization & dropout
+- Sigmoid output layer
+- Multi-label classification
+- Trained using Binary Cross-Entropy / Focal Loss
+
+---
+
+## 🚀 Inference Example
+
+```python
+import numpy as np
+from inference import predict_ecg
+
+ecg_signal = np.random.randn(1000, 12)  # replace with real ECG
+labels, probs = predict_ecg(ecg_signal)
+
+print("Predicted labels:", labels)
+print("Probabilities:", probs)

ecg_cnn_model.keras

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:816592cf1556aea1529da3d11c211089fb2832a5bef8d5fce4bbefbca5882b16
+size 49695056

inference.py

@@ -0,0 +1,68 @@
+import numpy as np
+import tensorflow as tf
+
+# Load trained model
+MODEL_PATH = "ptbxl_ecg_cnn_model.keras"
+model = tf.keras.models.load_model(MODEL_PATH, compile=False)
+
+# Diagnostic labels
+TARGET_NAMES = ["NORM", "MI", "STTC", "CD", "HYP"]
+
+# Default thresholds (can be replaced with tuned ones)
+DEFAULT_THRESHOLDS = {
+    "NORM": 0.5,
+    "MI": 0.5,
+    "STTC": 0.5,
+    "CD": 0.5,
+    "HYP": 0.5
+}
+
+
+def predict_ecg(ecg_signal, thresholds=DEFAULT_THRESHOLDS):
+    """
+    Predict cardiac abnormalities from a 12-lead ECG.
+
+    Parameters
+    ----------
+    ecg_signal : np.ndarray
+        Shape (1000, 12), preprocessed ECG signal
+    thresholds : dict
+        Thresholds for each class
+
+    Returns
+    -------
+    predicted_labels : list
+        List of predicted diagnostic labels
+    probabilities : dict
+        Probability per diagnostic class
+    """
+
+    if ecg_signal.shape != (1000, 12):
+        raise ValueError("ECG signal must have shape (1000, 12)")
+
+    # Add batch dimension
+    ecg_signal = np.expand_dims(ecg_signal, axis=0)
+
+    # Model prediction
+    probs = model.predict(ecg_signal, verbose=0)[0]
+
+    predicted_labels = []
+    probabilities = {}
+
+    for i, label in enumerate(TARGET_NAMES):
+        probabilities[label] = float(probs[i])
+        if probs[i] >= thresholds[label]:
+            predicted_labels.append(label)
+
+    return predicted_labels, probabilities
+
+
+# Example usage (for testing only)
+if __name__ == "__main__":
+    dummy_ecg = np.random.randn(1000, 12)
+    labels, probs = predict_ecg(dummy_ecg)
+
+    print("Predicted labels:", labels)
+    print("Probabilities:")
+    for k, v in probs.items():
+        print(f"{k}: {v:.3f}")

requirements.txt

@@ -0,0 +1,6 @@
+tensorflow>=2.13
+numpy
+scipy
+wfdb
+scikit-learn
+matplotlib