Quantum Machine Learning Improves Early Lung Cancer Detection

This research investigates the application of quantum-classical hybrid machine learning to enhance the early detection of lung cancer. The study specifically focuses on analyzing cell-free DNA (cfDNA) biomarkers, including fragmentomics and methylation patterns, which are known for their high-dimensional and nonlinear molecular signals. By encoding features into quantum Hilbert space using various feature maps and entanglement strategies, the team developed fidelity-based quantum kernels. These quantum kernels were then integrated with classical machine learning techniques like precomputed-kernel SVM and kernel-PCA logistic regression. The performance of these hybrid models was systematically evaluated against traditional SVM models trained on the original feature sets. The findings indicate that quantum-kernel models achieved competitive results across different datasets. Notably, for fragmentomics data, several quantum configurations with fewer features (20) demonstrated improved Area Under the Curve (AUC) relative to classical SVMs, suggesting their effectiveness in capturing complex, nonlinear cfDNA fragmentation structures. While classical SVM performed best for methylation data, quantum models remained competitive and sometimes enhanced specificity. The study concludes that quantum kernel methods hold promise for cfDNA-based lung cancer detection, particularly in leveraging intricate molecular signals.