Deep learning dataset will provide label-free segmentation of live cell images

Oct 14, 2021 | Shaoni Ghosh

Deep learning dataset will provide label-free segmentation of live cell images title banner

The upsurge of Deep Learning in the field of Artificial Intelligence enabled the former to venture into the technological realm and thus, furthering its evolution in the modern 21st century. It has accelerated the digital domain in terms of its excellence. 


Deep Learning, a revolutionary technique of Machine Learning, which is a subset of AI, has incredibly metamorphosed the field of technology to a great extent. 


(Must Check: Top 10 Deep Learning Applications)


The Research


Sartorius, the Life Science company is also an international pharmaceutical and laboratory equipment provider which deals with the segmentation of Bioprocess Solutions and Lab Products & Services.


Sartorius has recently open-sourced a deep-learning dataset named 'LIVECell' for label-free quantifiable compartmentalization of live cell images.


The Life Science Company has joined hands with the German Research Center for Artificial Intelligence (DFKI) in order to emphasize on how this dataset can be put to use in deep learning.


Neural Networks resolve the intricacies in the field of Artificial Intelligence, Machine Learning and Deep Learning. It comprehends the behaviour of the human brain which enables the computer algorithms to detect patterns. 


Neural Networks can prove their excellence in detecting cells. But in order to do that, they need proper training with high-quality datasets in order to grasp how to divide at its finest. 


(Also Check: Introduction to Neural Networks and Deep Learning)


The dataset comprises a set of 5000 label-free phase-contrast microscopy images. These images are composed of 1.6 million cells of eight-cell categories and are highlighted with well defined morphologies.


The images succeeded in showcasing a huge disparity with respect to the cell size and cell shape. This is due to the progression of cell growth from seeding densities in the primary stage to matured and fused form of monolayers.


According to the researchers, making use of a varied collection of cells along with confluence circumstances in the 'LIVECell' dataset would be able to assist deep-learning-related other segmentation models with more accuracy than it was done before.


Traditional vs Novel Approaches


Before, as reported by MarkTechPost, traditional image-based approaches necessitate "tedious customization" and "rigorous tuning" for several kinds of cells superimposed with varied morphologies.


Now, researchers have strengthened their modern methods with accuracy that could even train neural networks.


Before it was just restricted to just one kind of cell morphology. However, with the use of neural networks, they could manage multiple classes. This procedure would enable more "robust segmentation" and reduce "user-introduced biases."


There was a time when LIVECell data was not released and the researchers had access to a collection of label-free images of only 4,600 images retrieved from about 26,000 cells. And now, with the advancement in deep learning technology, the technological space has always opened up remarkably.

Tags #Deep Learning