Data Pre-processing Techniques for Brain Tumor Classification

Brain tumor detection and classification is a main concern owing to the global fatalities caused by it. Computer-aided design (CAD) techniques for classification of brain tumor are benefitting radiologists and doctors for error-free detection and correct prognosis. Convolutional neural network (CNN) is the most sought-after framework in deep learning for brain cancer detection due to its robust nature and efficient handling of large datasets. Pre-processing has a pivotal role in brain tumor classification framework architecture. Benchmark datasets acquired for training and testing of the CNN framework have to be pre-processed before being fed to the framework. Pre-processing techniques like data shuffling, resizing, normalization, and augmentation are done to enhance the image quality for its effective analysis. It also increases the reliability of the model by decreasing the learning time. The problem of underfitting and overfitting is also overcome by adhering to pre-processing techniques prior to feeding the dataset to the designed framework in deep neural nets. In this paper, benchmark dataset for brain tumor classification has been pre-processed in two different manners before being fed to the deep CNN model for the classification of brain tumor and results are compared.