Estimation of cerebral oxy- and deoxy-haemoglobin concentration changes in a layered adult head model using near-infrared spectroscopy and multivariate statistical analysis

The non-invasive measurement of cerebral oxy- (Delta HbO(2)(br)) and deoxy-haemoglobin (Delta HHb(br)) changes using near-infrared spectroscopy instruments is often affected by the absorption in the extracerebral layer. We have exploited the multivariate calibration (partial least squares, PLS) method to minimize the errors for a range of blood volume, oxygen saturation and extracerebral layer thicknesses. The changes in the mean time of flight of photons (Delta tau) and attenuation (Delta A) on the surface of a 3D adult head model were simulated using a finite-element method based on the diffusion equation. The PLS was then performed to identify the optimal number of detectors, their positions and weightings, to optimize the estimation of Delta HbO(2)(br) and Delta HHb(br). We define the 'nominal accuracy' as the accuracy of estimating Delta HbO(2)(br) and Delta HHb(br) over a nominal range of extracerebral layer thicknesses and 'robustness' as the accuracy beyond the nominal range. The results showed that for one or two detectors, Delta tau performed better than Delta A while using them together gave the best performance. When more detectors were used, the performances of using Delta tau, Delta A or both together became comparable, showing that a larger number of detectors can compensate for the performance of a simple Delta A measurement despite this measurement having a relatively lower sensitivity to intracerebral absorption changes.