The DCT-CNN-ResNet50 architecture to classify brain tumors with super-resolution, convolutional neural network, and the ResNet50
Anand Deshpande, Vania V. Estrela, Prashant P. Patavardhan
Abstract
Brain tumors' diagnoses occur mainly by Magnetic resonance imaging (MRI) images. The tissue analysis methods are used to define these tumors. Nevertheless, few factors like the quality of an MRI device and low image resolution may degrade the quality of MRI images. Also, the detection of tumors in low-resolution images is challenging. A super-resolution method helps overcome this caveat. This work suggests Artificial Intelligence (AI)-based classification of brain tumor using Convolution Neural Network (CNN) algorithms is proposed to classify brain tumors using open-access datasets. This paper hiders on a novel Discrete Cosine Transform-based image fusion combined with Convolution Neural Network as a super-resolution and classifier framework that can distinguish (aka, classify) tissue as tumor and no tumor using open-access datasets. The framework's performance is analyzed with and without super-resolution method and achieved 98.14% accuracy rate has been detected with super-resolution and ResNet50 architecture. The experiments performed on MRI images show that the proposed super-resolution framework relies on the Discrete Cosine Transform (DCT), CNN, and ResNet50 (aka DCT-CNN-ResNet50) and capable of improving classification accuracy.