Learning with minimal effort: leveraging in silico labeling for cell and nucleus segmentation

TL;DR

This paper introduces a pretraining approach using in silico labeling to significantly reduce the need for manual annotations in cell and nucleus segmentation tasks with deep learning.

Contribution

It proposes leveraging label-free microscopy images to predict fluorescent labels, improving segmentation performance with fewer annotated samples.

Findings

Reduces annotation requirements for segmentation models

Pretraining with in silico labeling enhances segmentation accuracy

Effective across various training set sizes

Abstract

Deep learning provides us with powerful methods to perform nucleus or cell segmentation with unprecedented quality. However, these methods usually require large training sets of manually annotated images, which are tedious and expensive to generate. In this paper we propose to use In Silico Labeling (ISL) as a pretraining scheme for segmentation tasks. The strategy is to acquire label-free microscopy images (such as bright-field or phase contrast) along fluorescently labeled images (such as DAPI or CellMask). We then train a model to predict the fluorescently labeled images from the label-free microscopy images. By comparing segmentation performance across several training set sizes, we show that such a scheme can dramatically reduce the number of required annotations.