Zero-Cost Deep Learning to Enhance Microscopy

Abstract

Combining microscopy image acquisition and deep learning improves image processing and analytics. However, deep learning requires knowledge of information technology and expensive hardware. Also, proper training of the network is essential for the successful prediction of unseen images, and understanding the limits of network training is important. The aim of this Master’s thesis is to make free deep learning tools accessible for users to use, learn and share these methods in the field of microscopy image analysis. We created user-friendly Google Colaboratory notebooks for microscopy image segmentation (StarDist), restoration (CARE), and denoising (N2V). These notebooks are an easy and free introduction to deep learning but the limited Graphical Processing Unit (GPU) provided inhibits large-scale use. This Master’s thesis is a part of a collaboration project called ZeroCostDL4Mic.