Quantum machine learning approaches applied to healthcare and well-being
Cargando...
Archivos
Fecha
2025-05-06
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad de Deusto
Resumen
Quantum computing (QC) differs from traditional computing by using quantum bits (qubits) and leveraging quantum phenomena like superposition, entanglement, and interference. These properties enable quantum computers to perform parallel computations, vastly surpassing classical systems. Quantum machine learning (QML) combines QC with machine learning to enhance data processing and classification. This thesis explores the theoretical advantages and practical applications of QML in healthcare. It begins with a summary of recent advancements, emphasizing QML's significance in the healthcare sector, followed by three key case studies demonstrating its practical use. The study adopts a systematic literature review approach, including research question formulation and article quality assessment using specific metrics. Of 2,038 records collected, 468 duplicates and 1,053 non-related to healthcare were excluded, leaving 55 articles from 2019 to 2024 for evaluation. The analysis provides a foundation for future research, promoting innovative solutions to healthcare challenges at the intersection of quantum computing and machine learning.
The first case study examines two QML algorithms Enhanced Quantum Support Vector Machine (E-QSVM) and Quantum Random Forest (QRF) applied to COVID-19 and influenza datasets. After data cleaning, normalization, and scaling, dimensionality is reduced by selecting the top 10 features using Chi-Square, ANOVA, and classical models for quantum efficiency. E-QSVM employs a parameterized quantum circuit with a ZZ feature map to capture complex data relationships, while QRF, implemented via Pennylane, uses two parallel QPUs, selecting the most accurate output. Results show E-QSVM and QRF outperform classical counterparts by 1\% and 2\%, respectively, with E-QSVM and QRF achieving up to 6\% improvement over recent quantum models for the COVID-19 dataset.
The second case study introduces a quantum version of the Fully Convolutional Neural Network (FCQ-CNN) for Ischemic Heart Disease (IHD) classification. The model integrates quantum and classical convolutional layers, a pooling layer, and a fully connected layer. Parameters are optimized using the Adam optimizer with a learning rate of 0.001 and a batch size of 32, while binary cross-entropy is used as the loss function over 100 epochs. Dimensionality reduction is achieved using SVM with Recursive Feature Elimination (SVM-RFE). Testing on 20\% of the dataset yields an accuracy of 84.6\% with a loss of 0.28. When compared to classical models (Optimized-CNN and FCNN) and previously published quantum models, the FCQ-CNN shows accuracy improvements of 8.6\%, 12.6\%, 3.5\%, and 1.8\%, respectively.
The third case study explores two quantum models: Quantum K Nearest Neighbor (QKNN) and Amplitude Encoding Variational Quantum Classifier (AE-VQC). Both models utilize amplitude encoding to transform input data into quantum state amplitudes and were evaluated on datasets with 8, 16, and 32 features, selected through a feature selection process. QKNN uses the quantum K minimum-finding algorithm to identify the nearest neighbors, measuring distances between quantum states using fidelity and applying a swap test for statistical assessment. AE-VQC comprises a feature map, variational quantum circuit, measurement circuit, and COBYLA optimizer. The experimental results demonstrate that both models outperform their classical counterparts, with QKNN achieving 3\%, 3\%, and 0.5\% higher accuracy across set of features. AE-VQC also shows improved accuracy of 0.3\% and 0.5\% in classifying sarcopenia disease, while reducing computational complexity.
In summary, this thesis evaluates various QML models, including data pre-processing and state preparation, across three diffrent case studies involving COVID-19, cardiovascular disease, and sarcopenia. The models show improved accuracy over classical machine learning techniques, highlighting QML's potential in addressing complex healthcare challenges efficiently. Despite these challenges like noise and limited resources, the findings contribute to the growing intersection of QML and healthcare, offering promising avenues for future research.
Palabras clave
Descripción
Materias
Matemáticas
Ciencia de los ordenadores
Inteligencia artificial
Ciencias Tecnológicas
Tecnología electrónica
Ciencia de los ordenadores
Inteligencia artificial
Ciencias Tecnológicas
Tecnología electrónica