Brain Tumor Classification and Severity Identification Using Deep Convolutional Spiking U-Net Lyrebird Neural Network and Alpha Piecewise Linear-Fuzzy

Purpose: To develop a robust framework that accurately classifies brain tumors and provides an estimation of their severity using an artificial intelligence approach to solve issues related to multimodal MRIs (Magnetic Resonance Imaging), such as resolution variability, misalignment and heterogeneity. Methodology: An intelligent Deep Convolutional Spiking U-Net Lyrebird Neural Network combined with an Alpha-Piecewise Linear-Fuzzy (CSULALF) model. Lyrebird Optimization (LBO) will improve parameter tuning and convergence of the Deep Convolutional Spiking U-Net. The Alpha-Piecewise Linear-Fuzzy logic will evaluate tumor severity through a combination of clinical indicators and imaging features. Findings: The proposed framework demonstrates improved performance over existing methods using traditional methods to handle multimodal inconsistencies and extract discriminative spatiotemporal features. The CSULALF model achieved an accuracy rate of 99.35% and a recall of 99.36%, demonstrating high capability for classification and prediction of severity. Research Limitations: The proposed framework exhibit decreased performance level for data that is skewed or noisy. There is an additional computational overhead associated with multimodal image processing and optimization methods, which could affect scalability. Practical Implications: The system enables reliable and automated clinical decision-making, improving diagnostic precision and supporting effective treatment planning in real world healthcare environments. Originality: This work presents a novel method that integrates spiking neural networks, Lyrebird Optimization, and Alpha-Piecewise Linear-Fuzzy, resulting in an explainable and unified approach to analyzing multimodal brain tumor imaging.