Hybrid Quantum Graph Neural Network for Brain Tumor MR Image Classification

Abstract

Brain tumor rank as the tenth leading cause of mortality among both adults and children. Early detection and treatment significantly enhance survival rates. Recent advancements in deep learning have demonstrated promise in identifying and classifying brain tumors using Magnetic Resonance Imaging (MRI) scans. This paper presents a novel approach that integrates classical and hybrid quantum-inspired graph neural networks for tumor classification. Classical Graph Convolutional Neural Networks (GCNN) analyse complex relationships within medical imaging data, while hybrid Quantum Graph Neural Networks (QGNN) improves performance by leveraging principles of quantum computing. Previous studies highlight the challenges posed by the diverse nature of brain imaging data. This study compares various classification approaches, emphasizing architectures, training techniques, and performance metrics. The objective is to train and evaluate models for identifying brain tumors from MRI scans. The hybrid QGNN demonstrates accuracy and loss metrics comparable to advanced classical GCNN, with accuracy improving from 0.41(41%) to 0.64(64%) during training and from 0.31(31%) to 0.52(52%) in validation datasets, thereby showcasing its effectiveness in distinguishing between normal and tumor images.