Hi Jim,
Pixel data are stored and organized as two-dimensional arrays in memory. In other words, you can think of an image as a matrix where the number of rows is the height in pixels of the image, the number of columns is the width, and each matrix element is a pixel value. As a result of this logical structuration, defining rectangular regions of interest with sides parallel to image borders is the best option performance-wise. Defining arbitrarily-shaped ROIs and previews is doable, but would incur significant performance penalties without a clear benefit in usability terms.
You can simply make a duplicate of the image, rotate it as desired, and define a preview on this rotated version. You can use this "rotated preview" as a ROI for white and black references, etc., and it can also be used to test processes. However, bear in mind that rotation implies interpolation, that is convolution, that is correlation among nearby pixels, and that can lead to misleading preview results in some cases.
