Figure 4.

Deep-learning approach performance on particle mixtures and in different environments. (a–d) Phase and amplitude signals from a representative particle from each characterized population. (e) Deep-learning approach (using 60 observations) distinguishes and correctly characterizes subpopulations in a multicomponent mixture dispersed in water, consisting of 0.21 μm silica (SiO₂), 0.15 μm PS, and 0.23 μm PS particles. Furthermore, the deep-learning approach accurately characterizes the radius r and the refractive index difference Δn of 0.23 μm PS particles dispersed in a 50% glycerol/water mixture, demonstrating that the measurements do not rely on detailed knowledge of the properties of the solution. The intersections of the dashed lines represent the expected positions of the populations. (f) When using a diffusion-based approach to characterize the multicomponent sample (dispersed in water), the size and refractive index distributions become very broad and the two smallest subpopulations cannot be distinguished when using 60 observations for the characterization.