Using transcutaneous spatially offset Raman spectroscopy (SORS) and partial least squares regression (PLSR), we recently predicted the areal bone mineral density (aBMD), volumetric bone mineralization density (vBMD) and maximum torque (MT) of tibiae in living mice. Despite the spatial offset geometry, the accuracy of the predictions was still affected by the signal from the overlying soft tissue that, like bone, contains large amounts of Type I collagen. Here we report a way to use SOLD (simultaneous, overconstrained, library-based decomposition) to improve the PLSR accuracy. The SOLD processing generates one bone spectrum estimate, one soft tissue spectrum estimate, and a residual. We combine the bone and residual spectra together for submission to PLSR, discarding only the soft tissue contribution. With the implementation of this soft-tissue-subtracted SOLD processing, we demonstrate that we can predict vBMD and MT more accurately than our previous transcutaneous measurements.
|