The characteristic features of a given scatterogram are first computed by projecting the pattern onto the Zernike orthogonal basis. Then the obtained data are compared to a database so that machine learning can yield identification result. A 10-fold cross-validation was performed on a database over 8 species (15 strains, 1906 scatterograms), at 6h of incubation. It yielded a 94% correct classification rate between Gram+, Gram- and yeasts. Results can be improved by using a more relevant function basis for projections, such as Fourier-Bessel functions. A fully integrated instrument has been installed at the Grenoble hospital’s laboratory of bacteriology and a validation campaign has been started for the early screening of MSSA and MRSA (Staphylococcus aureus, methicillin-resistant S. aureus) carriers.
Up to now, all the published studies about elastic scattering were performed in a forward mode, which is restricted to transparent media. However, in clinical diagnostics, most of media are opaque, such as blood-supplemented agar. That is why we propose a novel scheme capable of collecting back-scattered light which provides comparable results.