Augmented Multimodal Fusion for Optimized Brain Tumor Detection: Evaluation and Comparative Analysis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Brain tumors pose a significant medical challenge requiring accurate detection.
• Existing diagnostic methods may lack efficiency and timeliness.
• Advanced computational approaches are needed for improved brain tumor diagnosis.

Purpose of the Study:

• To propose and evaluate a novel brain tumor diagnosis method.
• To integrate and compare the performance of multiple pretrained base models.
• To establish a standardized technique for brain tumor detection model development.

Main Methods:

• Integration of pretrained models: VGG16, MobileNetV2, DenseNet121, InceptionV3, and ResNet50.
• Application of image augmentation techniques (brightness, contrast adjustment) for enhanced generalization.
• Utilizing data generators for efficient processing of large datasets.
• Systematic hyperparameter tuning and analysis using three optimizers (Adam, SGD, Adamax).
• Performance evaluation using metrics: accuracy, precision, recall, F1-score, and confusion matrices.

Main Results:

• MobileNetV2 demonstrated the highest performance with 96% accuracy, 96% precision, 94% recall, and 95% F1-score using the Adam optimizer.
• DenseNet121 and VGG16 achieved high accuracies of 95% and 94%, respectively.
• InceptionV3 and ResNet50 showed competitive but comparatively lower performance metrics.
• The proposed method proved robust and effective across evaluated models.

Conclusions:

• The integrated approach using pretrained models offers a viable and effective solution for brain tumor detection.
• MobileNetV2 is identified as a top-performing model for this specific diagnostic task.
• The standardized methodology facilitates model comparison and selection for clinical applications.