An interpretable deep learning framework for classifying neuronal morphologies using topology and graph neural networks

Imagine trying to sort a massive pile of unique, handcrafted trees. Some have branches that twist like pretzels, others stand tall and straight, and some spread out like umbrellas. In the world of neuroscience, these "trees" are neurons, and their shapes are crucial because a neuron's shape dictates how it processes information.

The problem is that sorting these neurons has been like asking a group of people to guess the breed of a dog just by looking at a blurry photo. Some experts say, "That's a Golden Retriever," while others say, "No, it's a Lab." It's subjective, inconsistent, and hard to prove who is right. Others have tried to measure specific parts of the tree (like the length of a branch), but that's like trying to identify a person only by their shoe size—it misses the bigger picture.

The New Solution: A "Super-Sort" Machine

This paper introduces a new, smart computer system designed to sort these neuronal "trees" fairly and clearly. Think of it as a detective that uses three different magnifying glasses to examine every neuron:

1. The Topology Lens: This looks at the "skeleton" of the tree. It ignores the exact curves and focuses on the big picture: How many main branches are there? Do they loop back on themselves? It's like looking at a subway map to see the overall structure of the lines, ignoring the specific stations.
2. The Graph Lens: This treats the neuron like a social network. It looks at how the branches connect to each other. Who is talking to whom? It maps the relationships between the different parts of the cell.
3. The Traditional Lens: This is the old-school method, measuring specific distances and angles, like a carpenter measuring a piece of wood.

How It Works

Instead of relying on just one of these views, the system combines all three. It's like having a team of experts where one is a map-reader, one is a relationship counselor, and one is a carpenter. They all look at the same neuron and vote on what type it is.

The paper tested this system against a group of human experts who had already labeled thousands of neurons. The result? The computer matched the humans' accuracy almost perfectly. This proves that to truly understand a neuron's "personality" (its type), you need to look at both its overall shape (the global structure) and how its tiny branches connect (the local details).

The "Why" Behind the "What"

Usually, deep learning computers are like black boxes: you put data in, and a label comes out, but you have no idea why the computer made that choice. This paper's system is different. It comes with a built-in "explanation manual."

Using a special tool called "explainable AI," the system can point to the exact branch or connection that made it decide, "This is a Type A neuron." It's like a teacher not just giving you the right answer on a test, but highlighting the specific sentence in the textbook that proves the answer is correct. This bridges the gap between the computer's math and the neuroscientist's understanding of biology.

The Bottom Line

The authors have built an open-source, transparent tool that sorts neurons objectively. By combining math, networks, and traditional measurements, they created a "Rosetta Stone" for neuronal shapes. This allows scientists to compare neurons from different labs or even different species with confidence, knowing they are all using the same, clear rules to define what makes a cell unique.