AI Learns to Trace Neuronal Pathways

Scientists at Cold Spring Harbor Laboratory (CSHL) have taught computers to recognize a neuron in microscopic images of the brain more than any previous approach. Researchers improved the efficiency of automated methods for tracing neurons and their connections, a task that is increasingly in demand as researchers work to map the brain’s condensate circuits. He did this by teaching computers to recognize different parts of neurons, each of which has different characteristics.

Such connection maps are important for learning how the brain processes information to generate thoughts and behavior.

In recent years, new imaging technologies and expanded ability to store digital images have given rise to a vast amount of data, capturing the paths of neurons as they make their way through the brains of mice and other model organisms Huh.

There are not enough experts to analyze all the images whose team has developed a new artificial intelligence (AI) tool, says CSHL professor Partha Mitra, and reported it in the journal Nature Machine Intelligence. Mitra says:

“I think the idea is to create a virtual neuranatomist for this project. And the reason for this is that the work we have been doing has traditionally been done by expert humans who, of course, require decades of training. He has a tremendous amount of knowledge.

They have seen – I don’t know – hundreds of thousands of images. They understand the context. And they can offer expert judgment and interpretation. ”

It should work with automated methods, says Mitra, but computers are not as good at interpreting visual information. An expert anatomist quickly identifies a single neuron as a crowded microscope image that is not as obvious as an algorithm — at least not without extensive training that allows computers to learn repeatedly from large datasets.

“Modern Machine Learning Techniques.” . . Still not very good. Mitra says, “We often do not have some knowledge or information about the knowledge or information that precedes us.” ‘”” So we need to build on some sort of prior notice. ”

Researchers using a form of mathematics called topological data analysis, a way to visualize it as 3D spaces with hills, valleys, and curves. The topology is sometimes called “ge rubber sheet geometry ‘which emphasizes connectivity,” Mitra says, in contrast, to the type of geometry that depends on the exact length and angle.

The researchers used simplified mathematical descriptions of the shape of neuronal parts — the plump cell body, slender axons, and branched dendrites. Neurons vary greatly in their overall shape, but the team greatly improved the program’s ability to detect axons and dendrites by showing the computer how neurons connect using some basic forms.

Mitra says, “Automated image analysis will still require a human proofreader for the future, to ensure quality in scientific applications – but by increasing the accuracy of computers, this new method significantly reduces the function that a Should be done by expert. ”

With the help of a new National Institutes of Health grant, the Mitra Group will develop its AI data analysis tool even more. These devices Are important for the Brain Initiative, of which his research is a part. He hopes that this approach will unravel the mystery of how brains connect so that humans can think about how brains actually work.

Semantic segmentation of microscopic neuroanatomical data by combining topological prizes with encoder-decoder deep networks

The understanding of neuronal circuitry in cellular resolution within the brain has relied on neuron tracing methods that include careful observation and interpretation by experienced neuroscientists. With recent developments in imaging and digitization, this approach is no longer possible with large-scale (terabyte to petabyte range) images. Machine-learning-based technology, using deep networks, provides an efficient alternative to the problem.

However, these methods rely on very large versions of annotated images for training and have error rates that are too high for scientific data analysis, and thus require substantial amounts of human-in-loop proofreading. Here we introduce a hybrid architecture combining the prior structure as topological data analysis methods, based on discrete Morse theory with the best-in-class deep-net architecture for neuronal connectivity analysis.

We show significant performance gains using its hybrid architecture over static structure detection (eg, connectivity of neuronal processes and local intensity maxima on axons corresponding to synaptic swellings on axons) closer to 90% compared with human observers miss.