Superresolution Imaging System Generates Segmented Fossil Images

Using an industrial grinder and superresolution camera, geoscientists have created 3D digital images of rock samples that can be viewed from multiple angles. Algorithms allow the computer to segment the images without human bias. Scientists used this technique to discover surprising information about Cloudina, an organism that lived over much of the world 545 million years ago and, until now, was thought to be one of the world’s first reef builders.

Scientists at Princeton University collaborated with SITU Studio to develop technology that can move through a rock and create digital renderings of more than a thousand wafer-thin slices of rock. The Princeton Grinding Imaging and Reconstruction Instrument, known as GIRI, allows geologists to see what rocks look like on the inside.

With an industrial grinder, a superhigh-resolution camera and high-speed neural networks, geoscientists can deconstruct rock samples and create 3D digital versions, which they have used to analyze specimens of Cloudina fossils gathered from the Byng Formation, a fossil reef formation in a glacier-carved valley on Salient Mountain in the Canadian Rockies. Courtesy of Adam Maloof and Akshay Mehra, Princeton University Department of Geosciences.

GIRI can cut slices as thin as a few microns. Each slice takes about 90 seconds to cut and image. Most of the specimens the scientists have imaged have been cut into 30-μm slices. A typical inch-thick, 1500-slice sample takes about a day and a half to grind and image.

GIRI can produce a 3D rendering of any solid object, regardless of whether it has the density differences needed for effective x-ray computed microtomography. A superhigh-resolution photograph is taken of every rock slice, so scientists always see the rock itself, not just a density model, such as provided by remote sensing.

The combination grinder and imaging system can create a 3D virtual tour through the inside of any solid object, from rocks to batteries. Here, a diamond wheel grinds a sample in the GIRI lab. Courtesy of Adam Maloof and Akshay Mehra, Princeton University Department of Geosciences.

Software algorithms developed by the team send and receive signals from the grinder, trigger the shutter, and verify image capture while GIRI is in operation. The team said that these machine-learning solutions make the process of image segmentation automated and reliable.

The researchers also designed a neural network image classifier to identify rock types by only their visual properties (color and texture). The network can predict which user-defined “class” — fossil versus matrix, for example — a pixel belongs to with greater than 90 percent accuracy.

GIRI pulverizes the sample, but it preserves every detail in its high-resolution images, which, according to its developers, arguably makes it less destructive than other approaches. It also preserves the structural information that reveals how the rocks formed.

“When people have attempted to get 3D information from rocks like this that are opaque to x-rays, they’ve always tried to dissolve the material out. But then you lose all the in situ information . . . this is a way to keep them (fossil specimens, for example) in their habitat while still trying to figure out what they looked like,” said professor Adam Maloof.

“You get to see photographs and make direct observations. That’s what’s been so life-changing to me: I love that it’s not a model . . . We’re on a virtual tour inside, rather than looking at waveforms and trying to interpret them,” he said.

Paleontologists, planetary scientists and engineers from around the world have reached out to the Princeton team to ask for virtual, visual tours inside their own specimens. From examining the delicate structures from the earliest life on Earth to investigating meteorites for hints to how the planets formed, “there’s really no limit to the contributions GIRI can make,” said Maloof.

The research was published in Proceedings of the National Academy of Sciences (doi:10.1073/pnas.1719911115).

This reconstruction of the Cloudina sample was created after Ph.D. student Akshay Mehra trained the neural network to distinguish fossils from surrounding stone, so he could produce this model showing only the fossil tubes. Courtesy of Adam Maloof and Akshay Mehra, Princeton University Department of Geosciences.