Integration of Electromagnetic Field simulation and Axial Scalable Context Adaptive Frilled Lizard Reverse Graph Attention networks framework for high-fidelity non-ionizing brain tumor imaging.

Abstract

The growing demand for safe and high-resolution medical diagnostics has accelerated the adoption of non-ionizing imaging modalities, particularly Electromagnetic Field (EMF)-based brain tumor imaging, which enables radiation-free visualization with enhanced tissue differentiation. Despite these advantages, conventional EMF imaging systems suffer from low sensitivity to weak signals, suboptimal reflection and transmission coefficients, and high computational complexity, leading to reduced diagnostic accuracy. To address these limitations, this paper proposes an integrated framework combining Electromagnetic Field Simulation with an Axial Scalable Context Adaptive Frilled Lizard Reverse Graph Attention Network (Axial-SCAF-LRGAN). The proposed model synergistically incorporates a Scalable and Adaptive Graph Neural Network and a Context Axial Reverse Attention Network to effectively capture contextual and spatial dependencies, while the Frilled Lizard Optimization algorithm is employed for optimal parameter tuning and enhanced learning efficiency. Experimental evaluations demonstrate superior electromagnetic performance, achieving a gain of 9.86 dBi, a reflection coefficient S11 of -37.42 dB, and a transmission coefficient S21 of -22.18 dB, indicating minimal reflection losses and efficient signal transmission. Computational analysis further reveals a low processing time of 0.29 s, RMSE of 2.41, and correlation coefficient of 0.99, reflecting high predictive consistency. The framework attains an accuracy of 99.8%, sensitivity of 99.7%, specificity of 99.6%, and F1-score of 99.75%, confirming its effectiveness for efficient, non-invasive brain tumor imaging with enhanced clinical applicability.