Yollanda-umutoni/Brain-tumor-classification-research

This repository hosts the code, trained models, and documentation from my Master's thesis: "Evaluating the Effectiveness of Deep Learning Approaches in Brain Tumor Classification using 2D MRI Scans" (University of Rwanda, African Centre of Excellence in Data Science, 2025). # Brain Tumor Classification using Deep Learning on 2D MRI Scans **Evaluating the Effectiveness of Deep Learning Approaches in Brain Tumor Classification using 2D MRI Scans** **Author:** Yollanda UMUTONI **Registration Number:** 219002673 **Degree:** Master of Data Science in Data Mining **Institution:** African Centre of Excellence in Data Science (ACE-DS), College of Business & Economics, University of Rwanda **Supervisor:** Dr. Gaspard GASHEMA **Submission Date:** October 2025 ## Overview This repository contains the implementation and resources for my Master's thesis, which compares three deep learning models **Custom CNN**, **ResNet50**, and **HiFuse** for multi-class brain tumor classification from 2D MRI scans. The study demonstrates excellent performance across all models, with **ResNet50 achieving 100% accuracy** on the test set, making it the most robust, stable, and clinically suitable model. The project aims to support faster, more accurate brain tumor diagnosis, especially in resource-constrained settings like Rwanda. **Keywords:** Brain tumor, MRI classification, Deep Learning, CNN, ResNet50, HiFuse, Medical Imaging, Transfer Learning ## Key Results | Model | Accuracy | Precision | Recall | F1-Score | Notes | |-------------|----------|-----------|--------|----------|------------------------------------| | Custom CNN | 99.84% | ~99.8% | ~99.8% | ~99.8% | Strong baseline | | HiFuse | 99.92% | ~99.9% | ~99.9% | ~99.9% | Good feature fusion | | **ResNet50** | **100%** | **100%** | **100%** | **100%** | Best overall – selected for deployment | - Near-perfect sensitivity and specificity - Very few (clinically insignificant) misclassifications - ResNet50 showed superior generalization and training stability Detailed classification reports, confusion matrices, training curves, and sample predictions are available in the Google Drive folder. ## Repos …