2.3.3. Proposed Framework

The generic structure of the isolated CNN that was developed from scratch is shown in Figure 1. Dataset-I and II were utilized for the training and testing of the 19-, 22-, and 25-layer isolated-CNN models. The 22-layer isolated CNN had the best accuracy for classifying brain MRI images into tumor and non-tumor class using dataset-I. The images of the non-tumor class of dataset-II were also used to train the binary-class isolated CNN. Finally, the pre-trained 2-class model was re-utilized using the transfer-learning method in order to re-adjust the weights of neurons to categorize the tumors into subclasses (glioma tumor, meningioma tumor, and pituitary tumor) for various tumor images of dataset-II. The complete framework of the proposed approach is shown in Figure 4.

The framework of the proposed approach.

The datasets I and II images were randomly distributed into the training and testing sets at the ratio 80 and 20%, respectively, to check the performance of the networks. For a fair comparison, all the parameters of training and validation were kept constant for each network.