Fig. 6: LodeSTAR measurement of particle polarizability exploiting signal strength symmetry.
From: Single-shot self-supervised object detection in microscopy
a Real and imaginary part of a simulated holographic image of a sphere (radius 228 nm, refractive index 1.58), (b) their versions with numerically rescaled signal strengths. c Despite being trained on a single particle (radius 228 nm, refractive index 1.58), the mean absolute percentage error (MAPE) of the predicted polarizability remains below 10% for a wide range of particle sizes and refractive indices. d In an experimental bi-dispersed sample, LodeSTAR accurately estimates the polarizability of the 150 nm population of polystyrene particles (pink histogram), even though it is trained on an image from the 228 nm population (orange histogram). e The imaginary part of a position-modulated holography image of fluorescent polystyrene particles (radius 225 nm) suspended inside and around SH-SY5Y human neuroblastoma cells imaged through an off-axis holography microscope. LodeSTAR, trained on a single image (bottom left), learns to detect and measure the fluorescent particles (orange markers) as well as the non-fluorescent intracellular particles (blue markers). See also Supplementary Movie 8. f The distributions of the measured polarizability of the particles and the biological particles are drawn from two distinct distributions, indicating that we successfully separate the added polystyrene particles from the biological particles. The peak of the distribution matches the expected polarizability of polystyrene inside of cells (dashed line), a less prominent peak near the expected polarizability of polystyrene outside of cells (dotted line).
