Storage can support medical research in many ways, especially as it relates to brain imaging hardware. How did Stanford University take advantage of high-speed storage in its recent research project?
In the medical field, fast storage enables researchers to store and parse large datasets. An example of a very large dataset is Stanford University’s MultiMAP project, which tracks, records, and analyzes the activity of nerve cells in the zebrafish brain.
What does Stanford’s MultiMAP do?
In a press release from Stanford University, science writer Bruce Goldman explained that MultiMAP (Multiple Alignment of Molecular and Activity Phenotypes) allows researchers to “track nearly every neuron in the zebrafish brain and then identify each associated The cell type of neurons.” The Stanford team behind MultiMAP used technology to determine that brain circuits are actually involved in alertness.
Stanford University has tracked neurons in the zebrafish brain, linking brain activity to alertness. Image via NICHD.
Karl Deisseroth, a professor of bioengineering and psychiatry and behavioral sciences at Stanford University and a Howard Hughes Medical Institute investigator, told Goldman, “We tracked every neuron in a fish’s brain while it was alive, when its cells were hyperactive, and learned to determine which cells were most active when the fish were most alert. Then, after preserving the fish’s brain tissue with a fixative without changing the relative position of the cells in the fish’s head, we could use molecular probes to find the relevant neurons and identify their cells Types of.”
The technology itself needs to operate at scale, so it requires a lot of low-latency, high-speed storage near the processing unit to function properly. As such brain imaging technologies continue to develop and advance, storage is also catching up.
MultiMAP is extremely important because the data it collects has a huge impact on mental health. In particular, Goldman noted that sleep deprivation and depression were both associated with a lack of vigilance, while anxiety and mania were associated with post-traumatic stress disorder (PTSD) and hypervigilance.
Deisseroth said: “The new findings open up a new avenue of in-depth exploration. The more we understand the neuronal architecture of brain states such as alertness, the more we will understand the concept of brain state itself, and we can even help design clinical trials for brain states. interventions.”
To embark on this “road of deep exploration,” fast storage is critical. Zebrafish larvae have approximately 100,000 neurons, and the data generated by each neuron needs to be recorded, stored, and combed through quickly and carefully to capture important data points. The adult zebrafish has about 10 million neurons, and the demands on Stanford’s MultiMAP technology have grown exponentially. By comparison, the human brain contains 86 billion neurons.
Humans have more than 860 times as many neurons as zebrafish.
What kind of storage can map the structure of the brain?
In processing so much information, mapping the brain’s structure and determining its decisions and functions, researchers are entering the realm of fast data, said Jason Echols, senior manager of technical marketing at Micron. In order to process so much data and analyze the important data points in it, MultiMAP “brings artificial intelligence and machine learning a little bit of clarity.”
“They need to do a lot of parallel processing,” Echols said. “When they need to bring data into the decision flow, they’re not just thinking, they’re actually taking the data and analyzing it.”
Medical researchers working on projects such as the MultiMAP project are struggling to “optimize storage infrastructure for optimal performance,” Echols said, especially when dealing with the massive amounts of data Stanford researchers have observed in brain cells.
“You’re going to find Nvidia and a few other companies working with them in these areas,” Echols said. “How they’ve made storage cost-effective and super-fast, with access to all the compute cores needed to run this data. What data is needed? That’s the real meaningful conversation.”
Powering all storage
Echols said storage research could power medical research. All aspects of this space will benefit as lean, fast and efficient storage solutions continue to be implemented in new applications.
Of the many ways that researchers and leading thinkers are applying memory and storage at scale, Echols explained, eight of these applications are advancing the frontiers of science and medicine:
Big Data and Analytics
Internet of Things (IoT) devices
Non-Volatile Storage Host Controller Interface Specification (NVMe) and Single Root Input/Output Virtualization (SRIOV)
Artificial Intelligence (AI) and Machine Learning
Software Defined Storage (SDS)
“Just looking at the storage side of the business, those are the areas where we can really impact at the macro-trend level,” Echols said.
Tony Ansley, chief technical marketing engineer at Micron, said many of these applications can benefit from the basic concept of thin storage: storing and computing data faster. For example, if researchers can tell whether a neuron is active or less active, they can draw conclusions more quickly.
“It’s not just a storage technology, it’s an architectural technology for the entire computer industry,” Ansley said. How to optimize storage for the needs of emerging server platforms, server technologies and processor technologies allows us to secure more in a smaller overall footprint. What about storage capacity and faster storage to improve analytics?”
Micron memory products such as Crucial DDR4 DRAM can help medical research technologies reduce processing time. Image via Dsimic.
Analyzing the data directly on the compute architecture takes about 5 nanoseconds to process. In terms of processing time, DRAM is extremely close to the computing architecture, taking about 30 nanoseconds. But as memory solutions move further and further away from the computing architecture, processing times also increase. A SATA SSD takes 300 microseconds to process the same data, while a SAS HDD takes 6 milliseconds. At the far end, a hybrid SAN takes a full 30 milliseconds.
That might not seem like a big difference, especially when the numbers are tiny fractions of a second, but Echols suggests explaining it in simple and understandable terms. If it takes 1 second to process data directly in the computing architecture, it takes 6 seconds for DRAM to process the same data. SATA SSD takes 16 hours, SAS HDD takes 2 weeks, and Hybrid SAN takes 2-3 months. At this level, Echols said, “These extremely fast computing architectures can’t wait that long.”
“That’s where Micron is innovating,” Echols said. “From DRAM to solid-state drives, we’re constantly innovating. That’s where we contribute to advancing the industry.”
the future of medicine
Ansley said one of the main ways Micron is helping to streamline medical research technology is by implementing storage solutions in a smaller form factor that researchers can take with them wherever they go.
“Especially in third world countries where there may be outbreaks of certain types of pathogens, researchers may need to collect real-time video scans or ultrasound scans, which is a lot of data to process, so they prefer to have the stored data in their hands,” Ansley said. rather than being pushed in a car.”
Tony Ansley believes that Micron’s memory products are ideal for supporting mobile medical devices. Image via Columbus Air Force Base.
Just as MultiMAP collects data from every neuron in the brain, other medical technologies collect vast amounts of data, from high-resolution X-ray images to gigabytes of data in ultrasound scan image files. Researchers need to have easy access to more data at any time in the field, avoiding constant uploads and downloads. So, Ansley said, “If I could integrate all the technology more efficiently into smaller and smaller packages, I would be able to hold it in my hand instead of carrying it on my back.”
He continued: “For example, Micron, as part of the industry, is always thinking about how to integrate more storage into a single NAND chip. We have expanded the storage of 2.5-inch HDDs or SSDs from 100GB in the past to 10 to 20TB, which is only a few short years away, and we will continue to work on it.”