Multi-Class Brain Tumor Type Classification Using Convolutional Autoencoder (CAE) With CNN-VGG19: An Experimental Study

The correct classification of brain tumors based on magnetic resonance imaging (MRI) is crucial in aiding clinical diagnosis and treatment planning. This paper suggests a multi-class brain tumor type hybrid deep learning model to classify MRI images into four classes, namely glioma, meningioma, pituitary tumor, and no tumor. The proposed method combines the use of a convolutional autoencoder (CAE) to extract the features of the image in an unsupervised manner and a transfer learning-based convolutional neural network (CNN) with the VGG19 architecture to classify the image finally. The CAE minimizes MRI images of high dimensions into small and discriminative feature representations, eliminating redundancy and maintaining important structural data. The resulting features are then fed into a fine-tuned CNN-VGG19 model with Global Average Pooling and fully connected layers with SoftMax activation for a 4-class classification. Data augmentation algorithms are used in order to alleviate overfitting and enhance generalization. It is experimentally shown that the proposed CAE combining the CNN-VGG19 model results in an overall classification accuracy of 92.39, which is better than the comparative CAE with the CNN-VGG16 models. The robustness and reliability of the model are further validated by evaluation based on micro and weighted precision, recall, and F1-score measurements. The results reveal that, in an MRI-based diagnostic system, the multi-class classification of brain tumor types in the case of unsupervised feature learning, in combination with transfer learning, is considerably improved.