Studying neural circuits of the brain is no easy task, not aided by the fact that there currently exists no viewable map of every single neuron, synapse, and other cells available.
But times are changing, and a team of scientists from Google and Harvard's Lichtman Lab has pooled its efforts together and has created a browsable 3D map of part of the cerebral cortex into the H01 dataset.
The team's efforts are a step forward in work Google and the Howard Hughes Medical Institute carried out last year, when they published the largest high-res image of brain connectivity of a fruit fly. The next logical, but tricky, step was the human brain.
The H01 dataset is based on one small section of the human brain, comprising imaging data that covers approximately one cubic millimeter of brain tissue, but it's one big step forward in creating a novel resource for studying the complex human brain, and improving and scaling the underlying connectomics technologies, as were always the team's goals.
And it was meticulous work. The dataset includes tens of thousands of reconstructed neurons, millions of neuron fragments, 130 million annotated synapses, 104 proofread cells, and a number of other subcellular annotations and structures — all of which can be viewed at leisure on the team's Neuroglancer browser interface.
In order to create their map, the scientists explain that they took a one cubit millimeter-big sample from the temporal lobe of a human cerebral cortex. After staining and coating it, the sample was cut into approximately 5,300 slices about 30 nanometers thick. These slices were imaged with a scanning electron microscope with a four nanometer resolution. This process created 225 million 2D images, which were meticulously stitched together to create one 3D volume.
Machine learning algorithms then stepped in to scan the 3D sample to determine its different cells and structures, which in the end lead to the H01 dataset — one of the world's most comprehensive maps of the human brain.
One of the most fascinating parts of the project is that the entire dataset takes up a whopping 1.4 petabytes of storage — roughly one million gigabytes. And that's just for one tiny section of the human brain.
The pre-print paper has been published in bioRXiv.