Synthetic neurons created by AI accelerate brain mapping.

Google researchers have developed AI technologies that use synthetic neurons to enhance the accuracy of brain mapping, which can significantly accelerate the process of reconstructing neural networks.

Artificial intelligence is beginning to play a key role in neuroscience, opening new horizons in the study of brain function. The work of Google researchers on synthetic neurons promises to speed up and improve the mapping of complex neural structures, which is particularly relevant given the challenges of creating brain maps for mammals and humans.

In the article "MoGen: Generating Detailed Neuronal Morphology through Point Cloud Matching," the team presented a model that uses synthetic neuronal shapes to enhance AI reconstruction models. Using machine learning methods, they were able to reduce reconstruction errors by 4.4%. While this may seem insignificant, it is crucial at the scale of the brain.

Brain mapping requires a massive amount of data and time. Researchers are analyzing the brain of a fly, which consists of 166,000 neurons, and are also actively working on reconstructing fragments of the brain of the zebra finch, the larva of the zebrafish, and a small section of the human brain. Recently, they began a mapping project of a small part of the mouse brain, which will be an important step forward.

The brain mapping process begins with creating three-dimensional images of neurons from two-dimensional slices of brain tissue. However, data processing requires significant human involvement. Researchers emphasize that properly functioning AI models can reduce the time spent correcting errors and make the process more efficient.

During experiments with the MoGen model, scientists investigated how increasing the volume of data, including synthetic samples, can enhance efficiency. The results showed that using synthetic neurons significantly improves accuracy in a fundamental way, opening new possibilities for large-scale neurobiology research.

Existing methods in neuroscience, such as MoGen and PATHFINDER, could radically change the approach to mapping and understanding the brain in the foreseeable future. This is a step towards creating in-depth and detailed models of neuronal activity, which will be of great significance for various fields of science.