A physics-guided neural framework for rheology measurement from dynamical laser speckles

Unlocking new frontiers in biomedicine and materials science requires rapid, non-destructive viscoelastic characterization of soft matter in its native state. Laser Speckle Rheology (LSR) has emerged as a transformative optical solution, fundamentally circumventing the invasive and time-consuming limitations of traditional rheometers. However, LSR in turbid fluids is severely constrained by multiple scattering, where standard physical inversions rely heavily on precise, sample-specific optical transport parameters difficult to measure in situ. To overcome this, this work bypasses the traditional Generalized Stokes-Einstein Relation (GSER) pathway by introducing a physics-guided deep learning framework that estimates an effective generalized Maxwell-mode representation from the intensity autocorrelation g 2 ( t ) and the speckle-intensity histogram. This effective representation is mapped through a fixed Maxwell forward model to predict G ′ ( ω ) and G ″ ( ω ) under physics-consistency constraints. This structured mapping reduces reliance on the ill-posed numerical transformations and error amplification associated with classical methods. Quantitatively, the framework achieves RMSE log as low as 0.009 against the reference and generalizes to previously unseen scattering conditions, preserving physically plausible frequency dependence and G ’( ω )– G ’’( ω ) phase behavior. By reducing reliance on these elusive parameters, it yields an interpretable, rheologically admissible effective Maxwell-mode representation, significantly improving the practicality of LSR in turbid media.