Fibonaccinet: A Biologically-Inspired Deep Learning Architecture for Accurate and Explainable Brain Tumor Classification From Mri

Brain tumor classification using magnetic resonance imaging (MRI) remains a critical yet challenging task in medical image analysis. In this study, we propose a novel convolutional neural network (CNN) architecture, termed FibonacciNet, which is inspired by the recursive and hierarchical nature of the Fibonacci sequence. The architecture integrates progressive contextual blocks (PCBs), attention mechanisms (CBAM), and multi-scale feature fusion to improve classification accuracy and computational efficiency. Furthermore, an advanced variant—Residual Path-Enhanced FibonacciNet (RP-FibonacciNet)—incorporates residual connections and the Mish activation function to enhance gradient flow and model generalization. Experiments conducted on benchmark datasets including the Brain Tumor MRI Dataset, BraTS, and Figshare confirm that our models outperform traditional architectures such as VGG16, ResNet50, and EfficientNet in terms of accuracy, AUC-ROC, and training stability. RP-FibonacciNet achieved a classification accuracy of 99.05% and a macro-average AUC of 99.78%. Additionally, feature map and attention visualizations demonstrated the model’s ability to focus on relevant tumor regions. This work highlights the promise of biologically and mathematically inspired neural networks in improving both performance and interpretability for clinical decision support in brain tumor diagnostics.