Estimating shear modulus of isotropic materials from scanning laser Doppler vibrometry via convolutional neural networks.

This study explores the use of Scanning Laser Doppler Vibrometry (SLDV) and Convolutional Neural Networks (CNNs) to estimate the stiffness of silicon-based materials. The research is motivated by the growing evidence that tissue mechanical property values are important parameters for diagnosis as they are sensitive to pathological changes. SLDV is a dynamic elastography technique that measures wave propagation and is non-contact, non-invasive, and relatively low-cost. CNNs have been shown to be able to assess mechanical properties from elastography images more accurately than traditional inversion techniques. Soft tissue-mimicking materials were used in the analysis to realistically simulate the properties of soft tissues, exhibiting similar deformation responses and stiffness values. Two different methods of mechanical vibration source were used to stimulate the specimens during imaging. The classification of the shear modulus of the materials was performed on two separate tasks: binary classification and a five-class classification. Open datasets of SLDV images were not present in accessible databases, so the proposed CNN architecture was pre-trained using synthetic wave data generated using a computational model and then fine-tuned with physical data. During the two experiments using physical data, the binary classification achieved an accuracy of 84.4%, and the multi-class classification reported an accuracy of 76.6%. While these results do not yet allow a clinical application for the estimation of the stiffness of organs and soft tissues, they constitute a step forward towards the implementation of an automatic and reliable method for assessing mechanical properties from elastography images.