Brain Tumor Classification Using a Controlled -Depth Convolutional Neural Network with Hyper Parameter Optimization

Authors

  • bara barati Assistant Professor of Medical Physics Author

Keywords:

Classification, Brain tumor, Deep learning, convolutional neural network

Abstract

Brain tumors pose significant diagnostic challenges due to their structural variability and clinical implication. In this study, we propose a convolutional neural network (CNN) architecture with a deliberately constrained depth, designed to achieve improved brain tumor classification while maintaining balanced model complexity and generalization performance. The model classifies MRI brain images four categories: meningioma, glioma, pituitary tumor, and no-tumor. The proposed architecture integrates sequential convolutional–pooling blocks followed by fully connected layers. A controlled design strategy was adopted to manage the number of trainable parameters and reduce overfitting while preserving feature extraction capability. The model was trained and validated on a dataset of 7,125 MRI images using a structured pipeline that included preprocessing (resizing, normalization, denoising, and augmentation) and supervised classification. Final performance evaluation was conducted on an independent test set (n = 1,425). The network achieved 99.96% training accuracy and 97.82% validation accuracy. On the independent test set, the macro-averaged accuracy was 97.71%, with precision (97.75%), recall (97.75%), and F1-score (97.50%). ROC analysis yielded an AUC of 1.0 for each class. Additional confusion-matrix-based metrics demonstrated high reliability (macro-specificity: 99.28%, MCC: 0.971, Cohen’s κ: 0.969), indicating consistent multi-class performance. The results demonstrate that a carefully controlled CNN architecture can achieve reliable and statistically supported performance for MRI-based brain tumor classification. Although further external validation is recommended, the proposed framework shows potential for supporting computer-aided diagnostic systems.

Downloads

Download data is not yet available.

Publication Timeline

Published

Issue

Section

Articles

How to Cite

barati, bara. (2026). Brain Tumor Classification Using a Controlled -Depth Convolutional Neural Network with Hyper Parameter Optimization. Journal of Artificial Intelligence, Applications and Innovations, 1-15. https://aiaijournal.com/index.php/aiai/article/view/89