It depends on the modality.
For cross-sectional imaging, the patient coordinate system relative
vectors in Image Orientation (Patient) are the most precise information,
and from them one can extract simple labels mathematically, so those
labels are not normally sent in Patient Orientation (since they would
be redundant).
For projection modalities, where the exact relationship to the patient
is not mathematically quantifiable but limited orientation is available
dependent on operator entry or assumptions about the geometry, the
labels in Patient Orientation are all that is available.
For one way to compute the labels from the vectors, see:
http://www.dclunie.com/medical-image-faq/html/part2.html#DICOMOrientation
In short, if Image Orientation (Patient) is present, use it, else use
Patient Orientation.
David