Artificial Intelligence (AI) for Brain Tumor Detection: Automating MRI Image Analysis for Enhanced Accuracy

Abstract

Accurately diagnosing and planning for the treatment of brain tumors is crucial in clinical practice. Brain tumor
detection and diagnosis rely heavily on artificial intelligence (AI) systems that mainly employ medical imaging
modalities like MRI. This study employs cutting-edge DL and image processing techniques to intelligently forecast the
brain tumor using AI. The complicated and varied nature of brain tumors frequently presents challenges to deep
learning models, despite their promising performance in this task. In order to overcome this obstacle, we present the
InceptionV3 architecture, which is based on CNNs and uses 5-fold cross-validation to classify brain tumors from MRI
images. A training, validation, and testing of a model were conducted using a publically accessible MRI dataset that
included 7023 greyscale brain MRI pictures. These images were classified into four types of tumors: gliomas,
meningiomas, no tumors, and pituitary. To enhance diversity of a training dataset, the photos were preprocessed by
scaling, greyscale conversion, and labeling. Afterward, data augmentation techniques were applied. A model's
performance was assessed using 5-fold cross-validation, yielding an F1-score of 99.98%, an average accuracy of
97.12%, precision of 97.97%, and recall of 96.59%. Other Artificial Intelligent models that were compared included
InceptionV3, VGG19, CNN, and DenseNet and the results indicated that the InceptionV3 gave better results overall.
These results demonstrate that deep learning can accurately and efficiently detect brain tumors utilizing MRI
pictures.