A high-throughput method for measuring fungal growth rate on solid media using automated imaging and deep learning

Abstract

Measuring the growth rate of filamentous fungi is an essential phenotype assay in fungal biology, enabling the comparison of nutrient-related fitness metrics across various isolates, species and genera. Conventional methods are time consuming and labor intensive, which prohibits the adaptation and implementation of high-throughput phenotyping. Here, we suggest a high-throughput methodological pipeline to study fungal growth on solid media combining the use of 24-well plates, an automated image acquisition system, and human assisted deep learning analysis of acquired images. Training a deep learning model through an iterative process - with continuous feedback and corrective annotations - enabled the development of a satisfying model that automatically segments pixels belonging to either fungus or background within a few hours. We evaluated this deep learning model by applying it to two test sets: First, a set of 336 images was used to validate the results by comparison with manual measurements. We demonstrate that the automated segmentation approach provides robust estimation of fungal growth not significantly different to manually segmented data. Second, a larger test set consisting of 2,016 images was used to illustrate the scalability of the model. After training the model for less than two hours, the deep learning model segmented the entire image data set automatically within minutes. The presented method is easily scalable and adjustable to other fungi and growth morphologies, due to the interactive training. Moreover, by combining 24-well plates and automatic image acquisition, measurements can be sped up as growth is detected across a smaller surface area than a standard six or nine cm diameter petri dish. The proposed methodological pipeline thus offers a new tool for estimating fungal growth rates, which can accelerate measurements, reduce bias, and increase throughput.