Partially interpretable image deconvolution framework based on the Richardson–Lucy model

Fluorescence microscopy typically suffers from aberration induced by system and sample, which could be circumvented by image deconvolution. We proposed a novel Richardson-Lucy (RL) model-driven deconvolution framework to improve reconstruction performance and speed. Two neural networks within this framework were devised, and they are naturally interpretable compared with previous deep learning methods. We first introduce RL into deep feature space and it has superior generalizability. We further accelerate it with a trainable unmatched backprojector, providing a five times faster reconstruction speed than classic RL. Our approaches outperform both convolutional neural networks and traditional methods in terms of image quality.