The Neural Network Attracted To The Cultivation Of Artificial Retina
Scientists have created a neural network-based algorithm that can monitor the formation of tissues inside the artificial retina and other organs and control this process better than a human, according to the MIPT press service, referring to an article in the scientific journal Frontiers in Cellular Neuroscience.
"The human retina has an extremely limited potential for regeneration; this means that any progressive loss of neurons, for example, in glaucoma, inevitably leads to complete blindness. Now doctors have almost nothing to offer such patients, except to start learning Braille tables.
Our work takes Biomedicine a step closer to creating cell therapy for the treatment of retinal diseases, which will not only prevent their development but also restore patients' already lost vision," explained Pavel Volchkov, one of the authors of the work, head of the MIPT genomic engineering laboratory.
Over the past ten years, scientists have repeatedly tried to grow artificial analogs of the liver and many other organs from embryonic or stem cells. The first such experiments were successfully completed back in 2012 – when Japanese scientists reproduced some of the liver functions of mice using stem cells.
As these experiments showed, the main problem in growing artificial organs was to make them form the correct layered structure that mimics the structure of real organs. Now scientists are achieving this by using three-dimensional polymer frameworks, as well as turning the patient's original organs, for example, one of the kidneys, into a kind of "template" for growth, clearing them of cells and leaving the proteins that frame them.
This can also be achieved by selecting a set of chemical signals that will cause the stem cells to form a so-called organoid-a miniature likeness of an organ. As shown by the experiments of the last five years, this approach allows you to create full-fledged analogs of retinal tissues, the brain, and a number of other complex organs.
The main problem with this technique, according to scientists, is that often stem cells turn into different types of tissue randomly, which, in fact, makes each organoid unique. This sharply limits their applicability in scientific experiments and clinical practice.
The victory of the neural network
You can deal with this problem by marking different types of cells with luminous proteins. However, because of this, organoids can no longer be used for transplantation. Russian researchers have solved this problem: their technique allows us to determine the number of different cell types and the structure of organoids using neural networks.
Unlike doctors and scientists who need to highlight the studied organoids, the neural network determines which cells are inside the organoid by analyzing its structure. Scientists tested its work on one of the most important tasks – predicting how artificial retinal tissue will form. Now doctors do not know how to restore it inside the body of patients, so if you learn to create artificial analogs of it, doctors will be able to return vision to their patients for the first time.
Volkov and his colleagues took a significant step towards solving this problem by obtaining photos of three thousand samples of growing artificial retinas, colored with fluorescent proteins, and without such "illumination". The researchers passed these images to colleagues who are good at distinguishing different components of the retina from each other and asked them to classify the photos.
The researchers then trained the neural network to identify cell types using 700 images and then tested it on 250 other photos. These experiments showed that the neural network coped with this task much better than a human – it received the correct answer in 84% of cases, while experts did it only for 67% of images.
"Our approach does not require complex images, fluorescent reporters, or dyes for analysis, it is easy to implement. This allows us to take another step towards creating cell therapies for retinal diseases such as glaucoma and macular dystrophy, which now almost inevitably lead to blindness," concluded Eugene Kegeles, an employee of the MIPT orphan disease therapy laboratory.