The Transformation Matrix is an implementation used for 3d graphics. It simplifies the math in order to speed up 3d positional / rotational orientations of points / objects. It is indeed very hard to pull out orientation from the Transformation because of the way it accumulates successive translations / rotations / scales.
Here's a suggestion. Take a vector that points in a simple direction like (1,0,0), and then apply the Transform to it. Your resulting vector will be translated and rotated to give you something like this: (27.8, 19.2, 77.4). Apply the Transform to (0,0,0), to get something like (26.1, 19.4, 50.8). You can use these two points to calculate the rotations that have been applied based on knowing their starting points of (1,0,0).
Does this help?