Polarization-resolved extension of Second Harmonic Generation microscopy (PSHG) exhibits proven efficiency in cancer diagnostics [1, 2]. Contrary to the case of white light microscopy, PSHG can reveal small structural collagen changes, during tumerogenesis, for a broad range of organs such as breast, thyroid, lung, pancreas, and ovary. However, despite its effectiveness for cancer diagnosis, PSHG is not yet fully exploited [1]. One way of improvement consists in taking better advantage of polarization-resolved measurements which are performed by acquiring multiple images (usually between 3 to 20) of the same sample under different input beam polarization conditions. Each image of the resulting stacked raw images set can contain relevant information not found in the other images of the set. In the literature, information extraction from stacked raw images is performed using methods such as averaging of all images [1], collagen structural parameters modeling [1] or PSHG polarimetric parameters extraction [2]. If the two latter methods provide a richer information than the first one, they may, however, suffer from a loss of information from the stacked raw images. To examine this potential loss of information, AI methods, see e.g. [3], can be used for extracting information from the stacked raw images. Using recently available images of the public SHG-TIFF database [1], dealing with breast and thyroid PSHG measurements of both normal and tumor tissues, we test available AI methods for information extraction and benchmark these methods to the stateof-the-art, in terms of automatic cancer diagnostic efficiency.