Abstract

Aim/Background: Early and accurate brain tumor diagnosis remains a major clinical challenge due to its impact on morbidity and mortality. Visual assessment of computed tomography (CT) and magnetic resonance imaging (MRI) can be affected by inter-observer variability and by the high volume and complexity of neuroimaging data. Therefore, computational tools that support specialists are needed to improve diagnostic accuracy and streamline neuroimaging workflows. This study proposes and evaluates a deep learning methodology for automatic brain tumor detection using multimodal CT and MRI data.

Methods: The proposed approach integrates multimodal fusion and advanced image processing in a modular pipeline comprising preprocessing, data augmentation (DA), and adaptation of the VGG16 convolutional neural network for binary classification (healthy vs. tumor-bearing brain). The model was trained and validated on a dataset available in the literature using 5-fold cross-validation (k = 5) and two experiments: (i) training on the original dataset without preprocessing or DA, and (ii) training following the full methodological sequence (preprocessing + DA + VGG16 adaptation) to enhance performance.

Results: The complete pipeline (with preprocessing and DA) improved model performance compared with training on the raw dataset, achieving high accuracy, sensitivity, and specificity in detecting tumor-bearing brains across the cross-validation folds.

Conclusion: The results indicate that multimodal CT–MRI integration combined with a VGG16-based classifier can provide robust automatic brain tumor detection, with potential to function as an auxiliary tool in clinical diagnosis. The proposed pipeline is modular and adaptable, supporting extensions to different clinical scenarios and future research on computer-assisted brain tumor detection.

Key words: Brain tumor; Multimodal imaging; Computed tomography; Magnetic resonance imaging; Deep learning; Fully convolutional networks