The general procedure of the HRL method can be adapted to phase unwrapping of other datasets. The general method is to (1) choose image statistics to use in the LDA, giving preference to those that will emphasize the edges of phase-wrapped regions (e.g., phase quality,41 fringe modulation,31 or edge detection filters34), (2) perform manual phase unwrapping for use as a training dataset, (3) perform LDA to separate edges of marked regions from all other pixels, (4) use image features as premarked inputs to RW segmentation,30 biased by the weights output by LDA, and (5) as an optional last step, tune LDA coefficients, for example, using a genetic search algorithm. We also note that different components of the HRL algorithm described here can be repurposed for other data processing tasks. For example, the combination of LDA plus a biased random walk would be an effective general method to segment images automatically, for example, to segment cells from background as an alternative to the widely used watershed algorithm.34 In this case, the LDA probability image would be used to combine multiple image features to enhance the precision of finding object or cell edges, eliminating the necessity for time-consuming manual image segmentation. As in the present study, a GA could then be used to further refine the classifier performance for the system of interest. Therefore, we expect that the HRL approach presented here is generally applicable to a variety of image classification problems, beyond phase unwrapping of biological QPI data.