High-performance visual geometric group deep learning architectures for MRI brain tumor classification

• Published in: The Journal of supercomputing Vol. 78; no. 10; pp. 12753 - 12764
• Main Authors: Veni, N., Manjula, J.
• Format: Journal Article
• Language: English
• Published: New York Springer US 2022; Springer Nature B.V
• Subjects: Accuracy; Artificial neural networks; Brain; Brain cancer; Cancer; Classification; Compilers; Computer Science; Deep learning; Image classification; Interpreters; Machine learning; Magnetic resonance imaging; Medical imaging; Mobile and Intelligent Sensing on High Performance Computing; Processor Architectures; Programming Languages; Deep learning; Magnetic resonance imaging; VGG architecture; Brain images; Classification
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-022-04384-9

Magnetic resonance imaging (MRI) of the brain is one of the most common imaging technologies used for brain cancer detection. Manual classification leads to more biopsies to ensure that there are no missed diagnoses. Recently, convolutional neural networks have achieved high accuracy in many image classification challenges. This study analyzes four different architectures from the Visual Geometric Group (VGG) for brain image classification using transfer learning. First, the feature space is generated using several 3 × 3 convolution filters and then reduced by the pooling layers in a block. These operations are repeated with different numbers of convolution filters in the subsequent blocks. After a predefined number of blocks, a fully connected layer is employed with an activation function to classify the given input. Four VGG architectures with different numbers of layers, 11, 13, 16 and 19, are developed to classify MRI images. The results prove that transfer learning on VGG architectures has good potential for brain cancer classification for REpository of Molecular BRAin Neoplasia DaTa (REMBRANDT) database MRI images. The results show that VGG-16 has the best performance, with an accuracy of 96% for brain cancer classification, followed by the VGG-19 architecture with 94.5% accuracy.