Human brain tumor classification using images of magnetic resonance and deep learning architectures
Year: 2025
Type: dissertation
Abstract: As the population and life expectancy around the world increase, diseases like cancer are becoming increasingly common. Research conducted by the National Cancer Institute indicates that in 2022 alone, more than 625,000 cancer cases were registered in Brazil, 11,000 of which were Central Nervous System (CNS) cancer. Early detection plays a fundamental role in treating this disease, and the most common way to diagnose brain cancer is through magnetic resonance imaging (MRI); however, the analysis of these images requires trained professionals. As a way to contribute to diagnosis, recent studies explore the use of image processing and Machine Learning algorithms to assist in the detection of human brain tumors. Considering this context, this work presents the creation of models based on convolutional neural networks (CNNs) and pre-trained architectures, ResNet and EfficientNet, to perform the automatic classification of different types of human brain tumors. Different configurations and architectures were evaluated, and the model that showed the best performance in the early detection of brain cancer was EfficientNet-B3, with an accuracy of 98.56%. However, regarding the balance of performance and accuracy, ResNet-50V2 proved superior, using less computational power while achieving an accuracy of 97.88%. For training the models, image pre-processing techniques were used in order to improve the model’s results and contribute to its final accuracy.
Author: Zorel, Pedro Lavarda
Language: Portuguese
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7155671230