Estimation of absolute wide-range blood flow by deep learning-based laser speckle contrast imaging

Laser speckle contrast imaging (LSCI) has been widely applied for blood flow measurements, while it is limited to obtain relative blood flow index (rBFI) rather than real blood speed in terms of mm/s due to its reliance on device parameters, static scattering noise and blood cells’ dynamic scattering type. Thus, the way that blood flow is estimated by LSCI model has not yet formed a standard, and its wide application is hindered, especially comparative experiments could hardly conduct among multiple conditions in biomedical researches. In this study, we developed a deep learning-based laser speckle contrast imaging model (DL-LSCI) to predict absolute blood flow through learning the distinct spatiotemporal frequency characteristics of speckle patterns from varying blood flow. Both phantom and in vivo experiments are performed. It demonstrates that our model enables to predict the blood flow within the widest, to best of our knowledge, absolute speed range of 0–231 mm/s under different static scattering noises in phantom experiments, as well as the varying blood flow in rat carotid artery occlusion/recovery model in vivo. The accuracies exceed 96 % and 92 %, separately. The study extends the LSCI device’s function in the absolute regional blood flow imaging free from the imaging system characteristics, extending the LSCI’s applications in studies where there is no baseline data and in comparative studies among animals. That would help to promote the LSCI technique becoming a standard quantitative imaging method to visualize the spatiotemporal evolution of blood flow in coronary artery bypass grafting.