Fig. 4.

Timing results and evaluation of necessary training samples. a Mean time for estimating oxygenation for a multispectral imaging stack of dimension $1000\times 1000\times 8$, where eight is the number of spectral bands used. Compared are the proposed random forest (RF) approach, support vector regression (SVR) and k-nearest neighbors (k-NN). b Median absolute error in oxygenation estimation as a function of the number of training samples. The results stabilize after training with about ${10}^4$ samples