Spatial and spectral regularization for multispectral photoacoustic image clustering

Photoacoustic imaging is a hybrid modality used to image biological tissues. Multispectral optical excitation permits to obtain functional images thanks to the tissue specific optical absorption that depends on the light wavelength. The aim of this study is to propose a clustering method for photoacoustic multispectral images based on both spatial neighbourhood and spectral behaviour. The proposed methodology is adapted from spatio-temporal mean-shift approach: it clusters distant or neighbouring patterns having similar spectral profiles. The clustering performance of our modified mean-shift algorithm is experimentally tested on multispectral photoacoustic tomography data. Results obtained from phantoms including two blood dilutions and colored absorbers are presented. It is thus shown that our strategy allows the experimental discrimination of media, achieving a clustering performance of more than 99%. Moreover, depending on the applied pre-processing the discrimination of different concentrations of a same medium is possible.