Hyperspectral dark-field microscopy (HSDFM) and analysis algorithms demonstrate classification of various tissue types, including carcinoma in human post-lumpectomy breast tissues. Performance of the HSDFM is evaluated by comparing the backscattering intensity spectra of polystyrene microbead solutions with Monte Carlo simulations of the experimental data. For classification algorithms, two approaches, a supervised spectral angle mapper (SAM) algorithm and an unsupervised K-means algorithm, are used. The manually extracted endmembers of known tissue types were determined by the histopathology reading of the hematoxylin and eosin (H&E)-stained slides. Their associated threshold spectral correlation angles from the SAM algorithm for supervised classification make a good reference library that validates endmembers from the unsupervised algorithm. For unsupervised classification, a K-means algorithm, with no a priori information, produces abundance maps with dominant endmembers of various tissue types, including carcinoma subtypes of invasive ductal carcinoma and invasive mucinous carcinoma. The endmembers extracted by the two methods agree with each other within less than 2% residual
|