Description

With a wide range of applications, high-throughput high-content microscopy has become an essential tool in biology and medicine. Today gigabyte-scale microscopy image datasets are routinely collected, stored, and analyzed. Despite significant advances made in fast imaging systems, collecting terabyte-scale data however is challenging. In such large-scale experiments, resolution is often sacrificed in favor of scale and time or vice versa. To address this problem, we propose Differentiable Microscopy , a design paradigm that optimizes optical systems and image processing algorithms together for higher throughput. In conventional microscopy image signals are acquired and often processed using algorithms such as deep neural networks. To extract information downstream, neural algorithms learn useful features (much smaller than the original data). In we will learn optical instruments to directly measure useful features needed for the downstream task. By doing so, we to measure images in their most information-rich form with minimum redundancy.