Fisher-information limits of detector-bandwidth-efficient 3D light-field microscopy

Abstract

Light-field microscopy enables snapshot volumetric imaging, but its information rate is constrained by both optical encoding and detector readout architecture. Here we develop a task-dependent Fisher-information framework that evaluates optical encoders relative to the detector resource limiting acquisition throughput. We compare full Fourier light-field microscopy (FLFM), squeezed light-field microscopy (SLIM), and frame-rate-matched FLFM under a common optical geometry, photon budget, and row-limited camera model. Sparse scenes are analyzed using a 3D point-emitter Fisher matrix, and dense scenes using Fourier-mode information on tilted spectral slices. At a compression factor of 0.25, SLIM provides 2.40× higher axial Fisher information per camera bandwidth and 1.89× higher 3D D-optimal position information than frame-rate-matched FLFM. For dense scenes, it provides 1.85× higher integrated Fourier-mode Fisher information per bandwidth, 4× greater axial-frequency extent, and approximately 11× larger projected lateral hard-support area. Sweeps over compression factor and view tilt show that these advantages reflect a general detector-allocation principle rather than a specific operating point. More broadly, the framework can be adapted to other camera architectures by incorporating architecture-specific measurement models and detector-throughput costs, providing a general basis for co-designing optical encoding, scene statistics, and camera readout.