Classification of brain tumours from MR images with an enhanced deep learning approach using densely connected convolutional network

Brain cancer is one of the most leading causes of death in human beings. There are different types of tumours affecting the brain and early diagnosis of them increases the survival rate. Classification of tumours from MR brain images is an essential task in treatment of the disease. Manual classification of tumours may lead to intra and inter observer variability and also time consuming. Hence, automated method of classification of brain tumours assists the doctors in diagnosis, classification and treatment of brain tumours. Since the past decade, deep learning based methods are widely used for classification problems especially in medical image classification. In this paper, an automated method of brain tumour classification is proposed based on enhanced deep learning approach using densely connected convolutional network (DenseNet). The transfer learning with DenseNet121 architecture is used for classification of brain tumours. The CNN model is optimised by tuning of hyper-parameters of the network, thereby improving the classification accuracy. The proposed method is evaluated on publicly available data set comprising 3064 MR brain tumour images belonging to three types of brain tumors - meningioma, glioma and pituitary tumours. It is inferred that DenseNet architecture gives better classification accuracy compared to VGG16, SVM and AlexNet. Through hyper-parameter tuning of the top dense layers of CNN, the classification accuracy improves by 5.26% for DenseNet121 architecture. The method performs superior compared to the state-of-the-art methods with a classification accuracy of 97.39%.