Enhancing Brain Tumor Classification Accuracy through Multi-Modal MRI Analysis and Advanced Ensemble Deep Learning
Loading...
Date
2025-06-30
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UNIVERSITY OF M’SILA
Abstract
In an attempt to apply state-of-the-art deep learning techniques for brain tumor classification,
which is a medical imaging domain, accurate multi-class diagnosis can dramatically
affect treatment choices and patient outcomes, two modern CNN architectures
EfficientNet-B3 and ConvNeXt-Tiny were evaluated for the classification of 44
different brain tumor types through Magnetic Resonance Imaging (MRI) data. The
dataset, sourced from Kaggle. To redress class imbalance and enhance the generalization
aspect, several data augmentation methods . Transfer learning was implemented
by employing pretrained weights, and both models were fine-tuned with customized
classifier heads that have Batch Normalization, Dropout and Dense layers for training.
The same hyperparameters were used to train all models, and the evaluation
was done with Accuracy, F1-Score, Recall, and Precision. The results demonstrate that
EfficientNet-B3 reached an accuracy of 96.85%, F1-Score 96.74%, Recall 96.85%, and
Precision 96.94%, while ConvNeXt-Tiny got accuracy of 95.88% ,F1-Score 95.86%, Recall
95.88%, and Precision 96.16%. A weighted ensemble approach was also used to
combine the two models. This ensemble model attained the highest classification accuracy
of 97.07%, which shows that weighted averaging is a good way to improve
performance. Such outcomes demonstrate the strong points of present deep learning
architectures and ensemble methods in dealing with fine-grained multi-class brain tumor
classification. The findings confirm that hybrid approaches have the capability to
surpass accuracy in addressing difficult diagnostic tasks from medical imaging data.