OxygenationBlood volume fraction Physics-informed models are important for understanding the photon propagation through the tissue and the spectra obtained in each pixel of the hyperspectral image.

GPU-DM is an inverse model developed for estimating tissue constituents in each pixel of the hyperspectral image by employing a two-layered diffusion model. Essentially is each pixel labeled with physical information like oxygenation, blood volume fraction, melanin amount, water content or other physical properties in a depth-resolved way.

Spectral fit in each pixel The independent nature of the employed model enables optimal GPU computing. Initialization of the model is done using gpudm_initialize(), which sets the model parameters according to a configuration file. A BIL-interleaved reflectance line is uploaded to the GPU using gpudm_reinitialize(). The parameters are fitted using gpudm_fit_reflectance(). Derived parameters can be downloaded to the host using gpudm_download_melanin(), gpudm_download_530res() and gpudm_download700res(), into new BIL-interleaved array with type of derived parameter along the band axis (oxyhemoglobin, deoxyhemoglobin, …). For a complete example, see main.cpp.

The model is described in Bjorgan et al. (complete citation: Bjorgan, M. Milanic, L. L. Randeberg, “Estimation of skin optical parameters for real-time hyperspectral imaging applications”, J. Biomed. Opt. 19(6) (2014)). A description is also available in this technical report.