Asad Channa

Pakistan

AI-assisted MRI-based classification of brain tumors for improved clinical diagnosis

Asad Channa¬π

¬π Department of Computer Science, Quaid-e-Awam University of Engineering, Science & Technology (QUEST), Nawabshah, Pakistan

Abstract

Background

Detecting brain tumors early is crucial for ensuring effective treatment and better outcomes for patients. However, diagnosing tumors from MRI scans can be slow and challenging, often relying on expert radiologists and being prone to human error. Differentiating between glioma, meningioma, pituitary tumors, and healthy brains can be especially tricky. In this context, artificial intelligence and in particular deep learning presents a powerful tool to help classify brain tumors quickly and accurately, potentially supporting clinicians in making faster and more reliable decisions.

Methods

We developed NeuroDetect, a deep learning model using VGG16 transfer learning to classify brain tumors into four categories: glioma, meningioma, pituitary tumor and no tumor. The model was trained on 7159 MRI images (glioma: 2,132; meningioma: 1,647; pituitary: 1,880; no tumor: 1,400) with stratified train-validation-test splits. Images were preprocessed by resizing and normalization and augmented with rotation, zoom and flipping to improve model generalization. The top layers of VGG16 were fine-tuned with dropout and batch normalization to prevent overfitting.

Results

The model achieved an overall accuracy of 94.4% on the test set, with high and balanced performance across all classes:

Glioma: precision 0.97, recall 0.94, F1-score 0.95

Meningioma: precision 0.91, recall 0.85, F1-score 0.88

Pituitary: precision 0.93, recall 0.99, F1-score 0.96

No tumor: precision 0.97, recall 1.0, F1-score 0.98

The confusion matrix confirmed that the model reliably distinguishes between tumor types, demonstrating its potential to support clinical decision-making.

Conclusions

Our findings show that NeuroDetect can accurately distinguish between different brain tumor types using MRI scans, suggesting that AI-based tools can support faster and more confident clinical assessment. Future work will focus on increasing the diversity of MRI data and testing more advanced deep learning models to strengthen reliability before clinical use.