Brain Tumor Classification using Deep Learning Framework

Brain tumor classification plays a prominent rolein accurate identification of abnormal brain tissues and helps in clinical diagnosis of patient. This work presents a brain tumor classification approach based on deep learning framework. Deep learning-based approaches have been used in the work due to its self-learning capability and outperformance in classification problems. In this work, to study brain tumor classification, 2D MRI data are used. The proposed method consists of three stages: i) pre-processing, ii) design of a deep learning architecture for brain tumor classification, and iii) integration of conventional handcrafted features with deep learning features. A detailed study has been done by training the proposed architectures with raw MRI images, Local Binary Pattern (LBP) coded texture features, and Discrete Wavelet Transform (DWT) coefficients for brain tumor classification. SoftMax classifier has been used for classification purpose. To authenticate the proposed method, publicly available brain tumor dataset (Br35H) has been used. The accuracies achieved are 81.11% when the proposed architecture is trained with LBP coded texture feature, 94.11% when network is trained with DWT coefficient, and 94% when raw MRI image is used for training ResNet architecture respectively at epochs 50. Further, the effectiveness of the proposed method is demonstrated by comparing its results with other existing methods. The experimental results of the proposed method show the efficacy of the proposed method over the methods considered for comparison.