A quantitative laser speckle-based velocity prediction approach using machine learning. (July 2023)

• Record Type: Journal Article
• Authors: Hao, Xiaoqi; Wu, Shuicai; Lin, Lan; Chen, Yixiong; Morgan, Stephen P.; Sun, Shen
• Is Part Of: Optics and lasers in engineering. Volume 166(2023)
• Publication Date: 2023-07
• Subjects: Laser speckle contrast imaging (LSCI) -- Blood flow imaging -- Machine learning -- Convolutional neural network (CNN) -- Microcirculation; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2023.107587
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: This manuscript proposed a laser speckle-based velocity prediction method using machine learning. Experimental data of rotating diffuser is adopted to train a three-dimensional convolutional neural network (3D-CNN) to establish a LSCI velocities prediction model (CNN-LSCI) with behavioral feature learning. The trained model has 0.33 MSE (mean square error) and 0.34 MAPE (mean absolute percentage error) and is verified by ten phantom velocities (0.2-4 mm/s, step is 0.445 mm/s) covering the typical blood flow velocity range of human body (0-2 mm/s) with the correlation of 0.98. The 3D-CNN is, for the first time to our knowledge, applied to LSCI for velocity prediction. Abstract: Laser speckle contrast imaging (LSCI) can be applied to non-invasive blood perfusion measurement with high resolution and fast speed. However, it is lack of measurement accuracy. The aim of this study is to enable quantitative measurement of LSCI by using Artificial Intelligence (AI), and this is achieved by using a set of experimental data obtained from a rotating diffuser (a tissue phantom mimicking blood flow under skins) within simulated flow velocity of 0.08–10.74 mm/s. These data were used to train a three-dimensional convolutional neural network (3D-CNN) to establish a LSCI velocities prediction model (CNN-LSCI) with behavioral feature learning. The trained model has 0.33 MSE (mean squared error) and 0.34 MAPE (mean absolute percentage error) and is verified by ten phantom velocities (0.2-4 mm/s, step is 0.445 mm/s) covering the typical blood flow velocity range of human body (0-2 mm/s) with the correlation of 0.98. The better performance of the proposed model is demonstrated by the results compared to traditional LSCI and multi-exposure laser speckle contrast imaging (MELSCI). This study shows the potential of LSCI to achieve quantitative blood perfusion measurement using machine learning.