Published Online:October 2025
Product Name:The IUP Journal of Computer Sciences
Product Type:Article
Product Code:IJCS011025
DOI:10.71329/IUPJCS/2025.19.4.43-66
Author Name:Kandagatla Srikar Prabhas, L Lakshmi and Abdul Talha
Availability:YES
Subject/Domain:Engineering
Download Format:PDF
Pages:43-66
Detection of brain tumors through image classification is a vital component of brain image processing, with the primary focus being identification of various brain tumor types such as meningioma, pituitary, glioma, and no tumor. This paper employs convolutional neural networks (CNN) and a transfer learning pretrained model known as visual geometry group (VGG)-16 model, specifically designed for the analysis of MRI brain scans. The dataset used for training these models comprises overall 7,023 images, in which 5,712 images are designated for the training set and 1,311 images for the testing set for validation. CNN model is trained on three unique input sizes, specifically 150 150, 192 192 and 224 224 pixels, resulting in achieved accuracies of 96.27%, 96.61%, and 98.25%, respectively. Conversely, the VGG-16 model is trained using images of sizes 192 192 and 224 224 pixels, which yielded accuracies of 97.22% and 97.32%, respectively. This comprehensive methodology proves highly efficient in the precise discovery of brain polyps, thereby enhancing medicinal care by enabling rapid identification of the four different classes of tumors from MRI brain scans. The integration of sophisticated pretrained architectures and transfer learning techniques has shown promising outcomes in medical image analysis, especially for detecting brain tumors. Using deep learning (DL) methodologies and preexisting models allows scientists to greatly improve the effectiveness and precision of classifying brain tumors. This progress ultimately aids patients by enabling quicker diagnoses and therapeutic actions.
The human brain, a vital component of the nervous system, serves as the command center for daily functions (Bassett & Gazzaniga, 2011).