A COMPARATIVE STUDY OF BRAIN TUMOR CLASSIFICATION USING DEEP LEARNING MODELS

Abstract

Brain tumor is a significant and often fatal disease, necessitating early detection for effective treatment. In this paper, we compare four deep learning techniques— DenseNet-121, ResNet-50, VGG-16, and Inception-V3—for classifying brain tumors using MRI images. The evaluation is based on accuracy, precision, F1-score, AUC ROC score, and Cohen Kappa score on the brain-tumor-detection-mri dataset from Kaggle, consisting of 2400+ images across two classes. Our results show an accuracy of 94.5% for DenseNet 121, 97.5% for ResNet-50, and 94% for both VGG-16 and Inception-V3, with a Cohen Kappa score of 73%. These findings provide insights into the strengths and weaknesses of each technique, aiding in the selection of the most suitable approach for medical image classification. Skin cancer is a prevalent disease worldwide, with increasing incidence rates. Early detection is crucial for successful treatment, as evidenced by statistics from the World Health Organization. In this proposed paper, we aim to develop a robust deep learning model for detecting benign or malignant skin cancer. We employ state-of-the-art pretrained deep neural network models, including Xceptron, EfficientNet, ResNet, and VGG-16, fine-tuned for our task. Our models achieve accuracies of 89.0%, 87.05%, 71.0%, and 83.33%, respectively. Additionally, we evaluate precision, recall, and F1-score for detailed analysis. The findings from our experiments are presented in this research work, offering valuable insights for skin cancer detection using deep learning techniques.