Medical images classification using CNN

Abstract

This report explores the application of deep learning for automated biomedical diagnosis through Convolutional Neural Networks (CNNs), integrating theoretical foundations with prac- tical application. It begins with a review of essential AI and deep learning principles, with par- ticular emphasis on CNN architecture and its effectiveness in feature extraction for both image and signal data. For the practical component, two dedicated models were implemented: a 1D- CNN to classify five types of cardiac arrhythmias from ECG signals in the MIT-BIH database, and a CNN to classify tumors from brain MRI scans. Following comprehensive preprocessing and training to address significant class imbalance, the models demonstrated high performance. The ECG classifier achieved an accuracy and F1-score of 0.99, while the brain tumor classifier achieved an accuracy of 0.91. These results highlight the robustness of deep learning models in biomedical analysis and their significant potential in supporting clinical diagnostic processes.