872 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 60, NO. 3, MARCH 2013 Multiscale Entropy Study of Medical Laser Speckle Contrast Images Anne Humeau-Heurtier*, Guillaume Mah´ e, Sylvain Durand, and Pierre Abraham Abstract—Laser speckle contrast imaging (LSCI) is a noninva- sive full-field optical imaging technique that gives a 2-D microcir- culatory blood flow map of tissue. Due to novelty of commercial laser speckle contrast imagers, image processing of LSCI data is new. By opposition, the numerous signal processing works of laser Doppler flowmetry (LDF) data—that give a 1-D view of microvas- cular blood flow—have led to interesting physiological information. Recently, analysis of multiscale entropy (MSE) of LDF signals has been proposed. A nonmonotonic evolution of MSE with two dis- tinctive scales—probably dominated by the cardiac activity—has been reported. We herein analyze MSE of LSCI data. We compare LSCI results with the ones of LDF signals obtained during the same experiment. We show that when time evolution of LSCI sin- gle pixels is studied, MSE presents a monotonic decreasing pattern, similar to the one of Gaussian white noises. By opposition, when the mean of LSCI pixel values is computed in a region of interest (ROI) and followed with time, MSE pattern becomes close to the one of LDF data, for ROI large enough. LSCI is gaining increased interest for blood flow monitoring. The physiological implications of our results require future study. Index Terms—Blood flow, laser speckle imaging, medical image processing, multiscale entropy (MSE), optical imaging, perfusion. I. INTRODUCTION T HE monitoring of microvascular blood flow has become a major interest in clinical routines and clinical research (see, e.g., [1]). Laser speckle contrast imaging (LSCI) is a re- cent noninvasive real-time imaging technique that has gained increased attention for a full-field imaging of microcirculatory blood flow [1]–[7]. LSCI exploits the random speckle pattern generated when tissue is illuminated by a coherent laser light. A camera is used to obtain a quick snapshot image of the time- integrated speckle pattern [8], [9]. The speckle size is deter- mined by the aperture size of the imaging device. With flow, the speckle pattern is decorrelated (blurred). The level of blurring is quantified by the speckle contrast value. A low level of blood Manuscript received February 28, 2012; revised July 2, 2012 and May 30, 2012; accepted July 9, 2012. Date of publication August 1, 2012; date of current version March 7, 2013. Asterisk indicates corresponding author. ∗ A. Humeau-Heurtier is with the Laboratoire d’Ing´ enierie des Syst` emes Automatis´ es, LUNAM Universit´ e, Universit´ e d’Angers, 49000 Angers, France (e-mail: anne.humeau@univ-angers.fr). G. Mah´ e and P. Abraham are with the Laboratoire de Physiologie et d’Explorations Vasculaires, LUNAM Universit´ e, Universit´ e d’Angers, 49033 Angers cedex 01, France (e-mail: gumahe@chu-angers.fr; piabraham@ chu-angers.fr). S. Durand is with the LUNAM Universit´ e, Universit´ e du Maine, 72085 Le Mans cedex 9, France (e-mail: sylvain.durand@univ-lemans.fr). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBME.2012.2208642 Fig. 1. Laser speckle contrast image (perfusion image) of a zone on the forearm of a healthy subject. flow leads to a high contrast; a high level of blood flow leads to a low contrast. The speckle contrast in the speckle pattern is used to create a perfusion map [1] (see an example in Fig. 1). The speckle contrast K is defined as the ratio of the standard deviation to the mean intensity I , as [10] K = σ s I  (1) where σ s refers to the spatial standard deviation of the speckle intensity. The speckle contrast depends on the exposure time T of the camera as [9] K 2 = σ 2 s (T ) I  2 = 1 T I  2  T 0 C t (τ )dτ. (2) C t (τ ) is the autocovariance of the intensity fluctuations in a single speckle C t (τ )= [I (t) −I  t ][I (t + τ ) −I  t ] t (3) where ...  t means a time-averaged quantity. However, it has also been shown that the second moment is properly calculated as [8] K 2 = σ 2 s (T ) I  2 = 1 T 2 I  2  T 0  T 0 C t (τ − τ  )dτ dτ  = 2 T I  2  T 0  1 − τ T  C t (τ )dτ. (4) Speckle contrast is computed from the mean and standard deviation of the speckle intensity which are generally deter- mined from a square region of N pixels , i.e., N 1 2 pixels × N 1 2 pixels . For large values of N pixels , the speckle contrast is estimated more accurately (the statistics are better) but the spatial resolu- tion is worse. For small values of N pixels , the large variation in the speckle contrast estimation reduces sensitivity to vascular variations. Rapid hardware devices for real-time laser speckle imaging have been proposed [11]. Cheng et al. [12] developed a technique where the contrast is calculated based on one pixel 0018-9294/$31.00 © 2012 IEEE