Classification and Segmentation of Brain Tumor MRI Images Using Convolutional Neural Networks

• Published in: 2023 IEEE International Conference on Engineering Veracruz (ICEV) pp. 1 - 6
• Main Author: Ruiz, Cesar Borja
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23-10-2023
• Subjects: Biological system modeling; Brain modeling; Brain tumor; Convolutional neural networks; Convolutional Neural Networks (CNNs); Deep learning; Image analysis; Magnetic resonance imaging; Magnetic Resonance Imaging (MRI); Medical image analysis; Predictive models; Training data
• DOI: 10.1109/ICEV59168.2023.10329651

The precise identification and segmentation of brain tumors play a crucial role in medical image analysis, enabling accurate diagnosis and effective treatment planning for a global challenge in central nervous system cancer. In this study, a deep learning-based model for multiclass brain tumor classification and binary segmentation in MRI scans is proposed. Based on convolutional neural networks, the model classifies four tumor types, achieving an overall accuracy of 93.56%. The segmentation model was oriented only to the meningioma class and the U -Net architecture was used. The model accurately delineates tumor boundaries, with a high similarity to the annotated masks under medical supervision. The segmentation model achieved an overall accuracy of 98.54 % and a mean intersection over junction of 81.96%, demonstrating its ability to handle complex tumor shapes and sizes. Both models were integrated into a web application capable of predicting new images and displaying the results. Extensive experiments were performed on the model hyperparameters and training data, and the results outperformed state-of-the-art approaches in accuracy.