Brain tumor classification of magnetic resonance images using a novel CNN-based medical image analysis and detection network in comparison with AlexNet.

Aim: This research work aims at developing an automatic medical image analysis and detection for accurate classification of brain tumors from a magnetic resonance imaging (MRI) dataset. We developed a new MIDNet18 CNN architecture in comparison with the AlexNet CNN architecture for classifying normal brain images from brain tumor images.

Materials And Methods: The novel MIDNet18 CNN architecture comprises 14 convolutional layers, seven pooling layers, four dense layers, and one classification layer. The dataset used for this study has two classes: normal brain MR images and brain tumor MR images. This binary MRI brain dataset consists of 2918 images as the training set, 1458 images as the validation set, and 212 images as the test set. The independent sample size calculated was seven for each group, keeping GPower at 80%.

Result: From the experimental performance metrics, it could be inferred that our novel MIDNet18 achieved higher test accuracy, AUC, F1 score, precision, and recall over the AlexNet algorithm.

Conclusion: From the result, it can be concluded that MIDNet18 is significantly more accurate (independent sample t-test <0.05) than AlexNet in classifying tumors from brain MRI images.