Implementing a Hybrid Quantum-Classical Neural Network by Utilizing a Variational Quantum Circuit for Detection of Dementia

TL;DR

This paper introduces a hybrid quantum-classical neural network using a variational quantum circuit for improved dementia detection from MRI images, achieving higher accuracy than classical methods.

Contribution

The study presents a novel hybrid neural network architecture that integrates a variational quantum circuit, enhancing MRI-based dementia classification performance.

Findings

QCCNN achieved 97.5% testing accuracy

QCCNN outperformed classical CNN in accuracy

High detection rates for normal and demented images

Abstract

Magnetic resonance imaging (MRI) is a common technique to scan brains for strokes, tumors, and other abnormalities that cause forms of dementia. However, correctly diagnosing forms of dementia from MRIs is difficult, as nearly 1 in 3 patients with Alzheimer's were misdiagnosed in 2019, an issue neural networks can rectify. Quantum computing applications This proposed novel neural network architecture uses a fully-connected (FC) layer, which reduces the number of features to obtain an expectation value by implementing a variational quantum circuit (VQC). The VQC created in this study utilizes a layer of Hadamard gates, Rotation-Y gates that are parameterized by tanh(intensity) * (pi/2) of a pixel, controlled-not (CNOT) gates, and measurement operators to obtain the expected values. This study found that the proposed hybrid quantum-classical convolutional neural network (QCCNN) provided 97.5% and 95.1% testing and validation accuracies, respectively, which was considerably higher than the classical neural network (CNN) testing and validation accuracies of 91.5% and 89.2%. Additionally, using a testing set of 100 normal and 100 dementia MRI images, the QCCNN detected normal and demented images correctly 95% and 98% of the time, compared to the CNN accuracies of 89% and 91%. With hospitals like Massachusetts General Hospital beginning to adopt machine learning applications for biomedical image detection, this proposed architecture would approve accuracies and potentially save more lives. Furthermore, the proposed architecture is generally flexible, and can be used for transfer-learning tasks, saving time and resources.