The utilization of AI technology enables accurate identification and classification of nerve cells in the brain, significantly saving time as well as cost.
FREMONT, CA – The human brain is one of the most complex organs in the universe, with hundreds of billions of neuron entangled in a sophisticated array. Every group of neurons is designed for a specific purpose and possesses varying shapes, sizes, and biological properties. The complexity of the brain is far beyond the reaches of modern science.
However, the incorporation of artificial intelligence (AI) in neurobiological analyses has enabled researchers to gain some insights into the brain structure. In this regard, the Max Planck Institute of Neurobiology and Google AI recently developed artificial neural networks to identify and classify the nerve cells individually based on their appearance.
The human brain comprises of around 86 billion nerve cells and about the same number of glial cells in addition to the 100 trillion connections between the nerve cells. Although mapping all the links is not possible, the researchers have developed a serial block-face scanning electron microscopy to target specific areas of the brain.
Studying even a 0.3 cubic meter section of the brain using an electron microscope requires several months since it comprises thousands of brain cells. Also, the collected data requires around 100 terabytes of storage capacity. Even more complex than the collection and storage is the data analysis of all the information.
The use of serial block-face scanning electron microscopy has dramatically enhanced the analysis techniques. The conventional methods required humans to identify and track the nerve cell connections using electron microscopes and conduct analyses. Hence, it often took years to develop even the tiniest data sets.
The implementation of AI enables the researchers to train the artificial neural networks to identify and classify the nerve cell components in the image data. It facilitates enhanced image analysis through flood filling networks, allowing the seamless extraction of image stacks without errors. However, the utilization of cellular morphology neural networks (CMN) will likely take the researchers a step further in the analysis. The human-like capabilities allow the cellular morphology networks to identify the cells by its shape and context rather than the image itself. Also, the CMNs can quickly identify the nerve cells in the image stack and classify it based on its appearance. It not only enables the researchers to understand the function of the cells but also the direction of information flow.