Project Description
This project addresses the critical need for early detection and classification of cardiac arrhythmias, leveraging deep learning techniques. Utilizing a convolutional neural network (CNN) with a sequential architecture, the model processes 2D spectral images of ECG data for accurate arrhythmia classification. The project demonstrates a scalable and practical solution for cardiovascular disease detection, achieving high accuracy while integrating user-friendly web application capabilities for real-time predictions.
Key Features
-
Data Preprocessing: Augmented and processed ECG data from six classes, including Normal and Ventricular Fibrillation, using techniques like rotation, flipping, and zooming for better generalization.
-
Model Architecture: Built a multi-layer CNN comprising convolutional, pooling, fully connected, and dropout layers to enhance generalization and reduce overfitting.
-
Performance Metrics: Achieved a high accuracy rate (~96%) with robust ten-fold cross-validation to demonstrate model reliability.
-
Application Development: Designed a web-based UI using Flask for users to upload ECG images and receive real-time predictions.
Technologies Used
-
Frameworks: TensorFlow, Keras
-
Programming Languages: Python (Flask for web integration)
-
Cloud Services: IBM Watson Studio for model training and deployment
-
Visualization: Matplotlib for visualizing model performance and prediction results
Outcomes
-
Early-stage detection of cardiac arrhythmias, enabling timely medical interventions.
-
Developed a user-friendly interface for clinicians and patients to classify arrhythmias with minimal effort.
-
Contributed to advancements in medical diagnostics by demonstrating a robust, scalable, and deployable machine learning pipeline.