Classification of Brain Tumor Images using Segmentation and Transfer Learning

• Published in: 2024 5th International Conference on Mobile Computing and Sustainable Informatics (ICMCSI) pp. 225 - 232
• Main Authors: Bindu, J. Hima, Devi, M. Uma
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18-01-2024
• Subjects: Brain tumor; Computed tomography; Computer architecture; Image segmentation; Magnetic resonance imaging; MRI; overfitting; segmentation; Training; Transfer learning; X-rays
• DOI: 10.1109/ICMCSI61536.2024.00040

Classifying brain tumors is considered imperative for developing successful treatments. In this regard, binary classification of brain tumor was the major objective of this paper. This work uses a magnetic resonance imaging (MRI) dataset for classification purpose. Due to data paucity, the classifying frameworks in this proposed work tested the transfer learning method for extracting features by utilizing four convolutional neural networks (CNNs) of pre-trained capabilities. This research aimed at decreasing the training time, improving accuracy in classifying images, and preventing overfitting. Instead of considering the same datasets, the study considered segmented images of the original dataset for training. Four segmentation methods were used for creating a new segmented dataset of the original dataset. The authors done segmentation and trained the architecture with few pre-processes for three separate epoch numbers in place for investigating the effect on classification accuracy and time consumption. Additionally, profits were gained from obtaining suitable results within a minimal number of epochs in a constrained time amount. The experimental data proved that the transfer learning method has produced accurate outcomes with a classification accuracy of 96.98%.