PT  - JOURNAL ARTICLE
AU  - Henry Pinkard
AU  - Zachary Phillips
AU  - Arman Babakhani
AU  - Daniel A. Fletcher
AU  - Laura Waller
TI  - Single-shot autofocus microscopy using deep learning
AID  - 10.1101/587485
DP  - 2019 Jan 01
TA  - bioRxiv
PG  - 587485
4099  - http://biorxiv.org/content/early/2019/03/23/587485.short
4100  - http://biorxiv.org/content/early/2019/03/23/587485.full
AB  - Maintaining an in-focus image over long time scales is an essential and non-trivial task for a variety of microscopic imaging applications. Here, we present an autofocusing method that is inexpensive, fast, and robust. It requires only the addition of one or a few off-axis LEDs to a conventional transmitted light microscope. Defocus distance can be estimated and corrected based on a single image under this LED illumination using a neural network that is small enough to be trained on a desktop CPU in a few hours. In this work, we detail the procedure for generating data and training such a network, explore practical limits, and describe relevant design principles governing the illumination source and network architecture.