STSA-Based Early-Stage Detection of Small Brain Tumors Using Neural Network

Early-stage brain tumor detection is critical for improving patient outcomes, optimizing treatment strategies, and enhancing healthcare resource allocation. However, existing state-of-the-art techniques struggle to detect tumors smaller than 5 mm due to their minimal dimensions and complex electromagnetic interactions. This study introduces a machine learning-based classification approach for early-stage Astrocytoma tumors (grades I and II) using step-constant tapered slot antenna (STSA) parameters. By leveraging scattering (S), admittance (Y), and impedance (Z) parameters as input features, an Artificial Neural Network (ANN) achieved a 99.95% classification accuracy for tumors with radii of 3 mm and 5 mm. Among the input features, impedance (Z) was identified as the most significant contributor to classification accuracy, whereas the S-parameter exhibited the lowest performance at 84.21% accuracy. The proposed methodology was benchmarked against Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest Classifier (RFC), and Graph Convolutional Neural Network (GCN), demonstrating superior classification performance across different tumor sizes. Additionally, the system maintained a low Specific Absorption Rate (SAR) of 0.30 W/Kg, reinforcing its suitability for biomedical antenna-based applications. An ablation study further confirmed that Z₂₂ and Z₁₄ phase components within the impedance matrix were particularly influential, as revealed through Local Interpretable Model-Agnostic Explanations (LIME), an explainable AI (XAI) technique. The proposed method was evaluated using a publicly available dataset, validating its robustness. These findings highlight the potential of STSA-based machine learning models for accurate, non-invasive early-stage brain tumor classification, enabling cost-effective, scalable diagnostics.