Enhancing Deep Learning with Quantum Computing: Exploring Hybrid Algorithms for Image Classification

Abstract

This thesis explores the integration of quantum computing and deep learning paradigms to enhance image classification, with a particular focus on detecting Alzheimer’s disease stages from MRI images. An innovative hybrid architecture, combining the pre-trained ResNet50 neural network for feature extraction with a variational quantum circuit (VQC) for classification, was developed and evaluated on the Augmented Alzheimer MRI Dataset. This approach leverages the hierarchical feature extraction capabilities of classical convolutional neural networks and quantum properties such as superposition and entanglement to enhance the representation of complex patterns. The results demonstrate an overall accuracy of 97.35%, with outstanding performance in discriminating moderate stages (F1-score of 99.92%) and confirmed robustness through low variance. A comparative analysis with state-of-the-art approaches positions the proposed model among the top performers, highlighting the potential of hybrid quantum-classical algorithms to overcome computational limitations of classical methods, despite the constraints of current Noisy Intermediate-Scale Quantum NISQ devices. This work also proposes perspectives for optimizing quantum circuits and extending this approach to other medical imaging tasks, emphasizing the importance of future advancements in quantum hardware for practical implementation.