KEYWORDS: Speckle, Blood circulation, Convolutional neural networks, Data modeling, Data acquisition, Microfluidics, Laser speckle contrast imaging, In vivo imaging
Laser Speckle Contrast Imaging is a well-established technique able to produce relative blood flow maps contactless and without using dyes. It relies on the statistical analysis of dynamic speckle images, observed when a coherent light is used to illuminate a medium that contains moving scatterers. The local speckle contrast is related to the movements of the scatterers. Multiple exposure speckle imaging (MESI) is a variant of the technique that takes advantage of multiple exposure data to retrieve more quantitative flow maps by accounting for the unwanted and superimposed contribution of static scatterers. Yet, in MESI, a model is adjusted pixelwise to the experimental data requiring long computation times and an a priori guess on the flow regimes. These issues hindered so far, the translation of MESI to clinical applications though some studies have already demonstrated its potential. Here we propose an alternative method based on Convolutional Neural Networks to analyze MESI data. The proposed CNN architecture has been trained and validated using experimental data acquired on calibrated microfluidics flow phantoms. Then, the trained network was applied to analyze MESI data acquired in vivo in mice brain. In addition to be model-bias-free, we have found that the CNN approach infers flow maps much faster than the classical pixelwise regression approach. This new approach is promising for the clinical translation of MESI.
Multiple Laser Speckle contrast Imaging (MESI) is an imaging method that provides relative blood flow maps from the statistical analysis of the dynamic speckle patterns observed when a coherent source is used to illuminate a tissue that contains moving scatterers. The gold standard analysis of MESI data is done by pixelwise regression of the experimental images to a theoretical function of the contrast K as a function of the exposure time T and decorrelation time τc. This approach is computer intensive, and the duration required to obtain a single flow map is too long for "real-time" analysis of in vivo hemodynamics. In addition, the mathematical model used relies on assumptions that oversimplify the local flow within the object of study. We have evaluated as an alternative a method based on Convolutional Neural Networks (CNN) to directly infer blood flow maps from MESI data, bypassing the model based fitting procedure. The CNN approach is model-free and delivers blood flow maps several orders of magnitude faster than the classical pixelwise non-linear regression. Here, we have evaluated two different datasets of annotated speckle contrast images to train the neural networks. One is composed of simulated time integrated speckle while the other one is composed of experimental data acquired for microfluidic channels with controlled geometries and flows. The study aims at discussing the assets and limits of both approaches.
Multiple exposure speckle imaging (MESI) allows to map relative blood flows at the surface of biological tissues. MESI is an extension of laser speckle contrast imaging (LSCI). It relies on the computation of speckle contrast K for several exposure times T, allowing to discriminate the contribution of static scatters (bulk tissues) and moving scatterers (red blood cells). The MESI model describes K(T) as a function of tc, rho, beta, and v. These variables are respectively the decorrelation time of the moving scatterers, the relative contribution of static scatterers to the speckle pattern, a normalization factor for the imaging parameters and the contribution of noises to the speckle contrast. In LSCI theory, tc is commonly assumed to be inversely proportional to the flow. The acquisition of the speckle data at multiple exposures and the subsequent non-linear fit on a pixel-wise basis are instrumentally complex and time-intensive tasks that prevent real-time computation of the flow maps. In the study, we evaluated the feasibility of machine learning analysis of MESI data to bypass the non-linear fitting procedure based on the synthetic exposure acquisition. Synthetic exposures limit acquisition bias due to imperfect illumination normalization and are less sensitive to camera noises except for low illumination conditions or imaging of fast flows. A residual convolutional neural network was adopted to predict the blood flow map based on a database of representative speckle images of channels in a microfluidic chip with calibrated flows. The MESI database contains images with different exposure times for different flow and different channel diameters. The database was spitted into a training and testing data set with a 50:50 ratio. Preliminary results showed that blood flow mapping using deep learning can achieve moderate accuracy and yield a more stable prediction with high noise-resistant ability, compared to pixel-wise non-linear fit.
Multiple exposure speckle imaging (MESI) allows to map relative blood flows at the surface of biological tissues. MESI is an extension of laser speckle contrast imaging (LSCI). It relies on the computation of speckle contrast K for several exposure times T, allowing to discriminate the contribution of static scatters (bulk tissues) from that of moving scatterers (red blood cells). First, we have evaluated how a synthetic exposure acquisition scheme could strongly simplify the instrument for MESI, while remaining quantitative over a range of relevant flows. A microfluidic chip with controlled flows in channels with dimension representative of mice brain cerebral vasculature has been imaged using the classical modulated intensities approach and the synthetic exposure mode. This study allowed to propose guidelines in terms of readout dark noise and spatial response uniformity for the choice of a camera for MESI in the synthetic exposure mode. Second, we have evaluated how unwanted movements introduce bias in the speckle contrast calculation for a representative range of movement speeds. Mixed solutions of intralipid and glycerin in Brownian motion have been characterized to provide calibrated samples in terms of scatterers de-correlation times. High concentration of glycerin led to decorrelation times of several ms corresponding to actual values in small capillaries while low concentration of glycerin led to decorrelation times of 1ms or less corresponding to arterioles and arteries. The effects of the unwanted movement speed and direction have been measured for both lateral (x-y) and axial (z) movements. The bias introduced by unwanted movement in the (x-y) plane depends on the relative values of the time between frames and the scatterers decorrelation. In addition, for axial movements, parameters such as the numerical aperture (NA) and the magnification level (M) need to be considered due to their role in defining the depth of field.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.