OpenCL™ framework for fast estimation of optical properties from spatial frequency domain images

摘要

To overcome the drawbacks of the commonly used lookup table inverse models, we propose a novel custom OpenCL™-accelerated artificial neural network inverse model for spatial frequency domain imaging (https://bitbucket.org/xopto/rftroop). Utilizing a mid-range graphics processing unit, the proposed inverse model can estimate high-definition (1920× 1080) maps of the absorption and reduced scattering coefficients and two scattering phase function related quantifiersat a rate of more than 50 frames per second. We show that the artificial neural network inverse model can be seamlesslyextended to estimate multiple optical properties independently, thus providing a versatile framework that allowsintroduction of new quantifiers.