AboutDownloadsDocumentsForumsIssuesNews

This project is an image dataset submission for a study which will be submitted for publication to PLOS One Computational Biology. The dataset contains a training set of 1086 images, a validation set of 274 images as well as several test sets, all in all


This project is an image dataset submission for a study which will be submitted for publication to PLOS One Computational Biology. The dataset contains a training set of 1086 images, a validation set of 274 images as well as several test sets, all in all containing >760,000 bacteria across 5 genera. Each image has a labelled corresponding mask which indicates which pixels belong to cells as well as which genus the cell belongs to, thus it is suitable for pixel classification tasks. The author summary which will be submitted to PLOS One is below:

"Microscopy is an invaluable tool when working with microorganisms. The morphology of bacteria can be highly indicative of its health as well as its genus, thus microscopy or characterizing the morphology is a routine task in the medical and biotechnological fields. However, evaluating the images requires specialty expertise from operators and has traditionally been time-consuming to analyse digitally. This study presents an automated and unsupervised method based on convolutional neural networks which outputs the basis of converting the morphology seen in the images, into digital datasets, by providing segmentation masks in which cells can be measured. Furthermore, these masks can be used to measure fluorescence images, further digitalizing signals from imagery. Most of all, the method is capable of classifying the genus of bacteria at a cell-level in a dataset comprised of 4344 images of cells from five genera of lactic acid bacteria. Implementing software such as this can vastly increase the value created by industrial screening workflows, in part by allowing the digitalization of cellular morphology for further processing or inclusion in other machine-learning algorithms. The classification of bacterial genera could be applied in automated contamination detection systems in manufacturing processes or imaging of samples containing multiple genera."

Feedback