Non-contact rheological assessment of hydrogels using deep learning and laser speckle imaging

Abstract

In this study, deep learning techniques are explored to derive the viscoelastic properties of samples from time series speckle image data obtained through laser speckle imaging. Rheological properties are inferred from the speckle patterns generated by the interaction between coherent light and the microstructure of the material. For samples with different viscoelastic modulus, corresponding temporal and spatial variations in speckle patterns are observed. In this paper, deep learning models including 3DCNN, CNN-LSTM, ConvLSTM, and SwinLSTM were implemented to predict viscoelasticity levels from laser speckle images of different hydrogel samples and extract both spatial and temporal features from the data. These models were trained to predict the viscoelastic modulus of hydrogel samples and validated with mechanical measurements. Comparative performance analysis between models showed superior results in a multi-task training using CNN-LSTM models on laser speckle imaging data. This study suggested that well-designed deep learning models can improve the accuracy and efficiency of laser speckle imaging-based rheological measurements, offering significant potential for non-invasive, real-time assessment of mechanical properties of biological tissues and soft materials.