Choice of numerical implementation of spatial contrast calculation impacts microcirculation quantitation in laser...

Laser speckle contrast imaging (LSCI) allows noninvasive imaging of microcirculation. Its scope of clinical applications is growing, yet the literature lacks a comparison of the accuracy of methods used to compute the spatial contrast $K_s$ from which the blood flow index is derived.

We aim to evaluate the impact on flow quantitation of different computational approaches used to derive $K_s$ .

We compare numerical calculation of $K_s$ in Python and ImageJ applied to noise-free simulated data and to experimental data acquired in vivo in anesthetized mice. The estimation of the decorrelation time ${\tau }_c$ , inversely proportional to the blood flow index, is carried out following two approaches: LSCI asymptotic estimation and fitting the multiple exposure speckle imaging (MESI) model to $K_s\left(T\right)$ .

For simulation data, we found variations of up to 58% for the blood flow index in the LSCI approach. Nonlinear fitting of the MESI model was less affected with discrepancies of only a few percent. Considering experimental data, the LSCI approximation led to $K_s$ with relative differences (up to 35%) depending on the calculation methods. The noise and limited exposure time strongly limited the accuracy of the LSCI asymptotic estimation. Adjustment of the MESI model to the data led to consistent values of ${\tau }_c$ in the 0.05 to 1 ms range with significant variations depending on the method used to calculate $K_s$ .

Numerical methods used to calculate $K_s$ should be precisely acknowledged and validated against direct calculation to ensure accuracy. Uniform filter approach leads to accurate $K_s$ values and is 100 times more computationally efficient than the $Direct$ calculation. Other investigated methods lead to various levels of errors in flow index estimation using LSCI. Errors are minimized using larger kernels. MESI derivation of ${\tau }_c$ is not immune but less affected by such methodological biases.