Quantum machine learning enhanced laser speckle analysis for precise speed prediction.

Area of Science:

• Biomedical Optics
• Quantum Computing
• Machine Learning

Background:

• Laser speckle contrast imaging (LSCI) is crucial for assessing blood flow perfusion but faces limitations in quantitative analysis.
• Three-dimensional convolutional neural networks (3D CNNs) improve LSCI's quantitative performance by extracting spatiotemporal features.
• Excessive downsampling in 3D CNNs can lead to critical information loss, hindering accurate blood flow quantification.

Purpose of the Study:

• To develop a hybrid quantum-classical 3D CNN framework to enhance the quantitative performance of LSCI.
• To address information loss in traditional 3D CNNs by integrating variational quantum circuits (VQCs).
• To improve the accuracy and stability of blood flow velocity prediction in LSCI.

Main Methods:

• Proposed a hybrid quantum-classical 3D CNN framework utilizing variational quantum algorithms (VQAs).
• Replaced the 3D global pooling layer with variational quantum circuits (VQCs) for enhanced feature integration.
• Validated the framework through cross-validation on experimental LSCI speckle data and evaluation on an unseen test set.

Main Results:

• Hybrid models demonstrated superior prediction accuracy and learning stability compared to classical 3D CNNs.
• Achieved up to 14.8% improvement in mean squared error (MSE) and 26.1% in mean absolute percentage error (MAPE) on an unseen test set.
• Qualitative analysis confirmed substantial improvements in predicting both low and high blood flow velocities.

Conclusions:

• The hybrid quantum-classical 3D CNN framework significantly enhances LSCI's quantitative capabilities for blood flow analysis.
• VQCs effectively preserve spatiotemporal information, leading to more powerful learning and generalization.
• This approach offers a promising direction for advancing quantitative optical imaging techniques.