System Bits: June 19

ML algorithm 3D scan comparison up to 1,000 times faster
To address the issue of medical image registration that typically takes two hours or more to meticulously align each of potentially a million pixels in the combined scans, MIT researchers have created a machine-learning algorithm they say can register brain scans and other 3D images more than 1,000 times more quickly using novel learning techniques.

MIT researchers describe a machine-learning algorithm that can register brain scans and other 3-D images more than 1,000 times more quickly using novel learning techniques.
Source: MIT

They explained that medical image registration is a common technique that overlays two images, such as magnetic resonance imaging (MRI) scans, to compare and analyze anatomical differences in great detail. For example, if a patient has a brain tumor, doctors can overlap a brain scan from several months prior onto a more recent scan to analyze small changes in the tumor’s progress.

The algorithm works by ‘learning’ while registering thousands of pairs of images, and in doing so acquires information about how to align images and estimates some optimal alignment parameters. After training, it uses those parameters to map all pixels of one image to another, all at once. This reduces registration time to a minute or two using a normal computer, or less than a second using a GPU with comparable accuracy to state-of-the-art systems, the researchers said.

Guha Balakrishnan, a graduate student in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Department of Engineering and Computer Science (EECS), and co-author on two new papers on this topic said, “The tasks of aligning a brain MRI shouldn’t be that different when you’re aligning one pair of brain MRIs or another. There is information you should be able to carry over in how you do the alignment. If you’re able to learn something from previous image registration, you can do a new task much faster and with the same accuracy.”

The VoxelMorph algorithm is powered by a convolutional neural network (CNN), a machine-learning approach commonly used for image processing. These networks consist of many nodes that process image and other information across several layers of computation.

The researchers noted that the algorithm has a wide range of potential applications in addition to analyzing brain scans. MIT colleagues, for instance, are currently running the algorithm on lung images.

They also believe the algorithm could pave the way for image registration during operations. Various scans of different qualities and speeds are currently used before or during some surgeries. But those images are not registered until after the operation. When resecting a brain tumor, for instance, surgeons sometimes scan a patient’s brain before and after surgery to see if they’ve removed all the tumor. If any bit remains, they’re back in the operating room.

With the new algorithm, the team said surgeons could potentially register scans in near real-time, getting a much clearer picture on their progress. Today, they can’t really overlap the images during surgery, because it will take two hours, and the surgery is ongoing but if it only takes a second, you can imagine that it could be feasible.

Low-voltage actuator for soft, wearable robotics
In the world of robotics, soft robots are the new kids on the block. The unique capabilities of these automata are to bend, deform, stretch, twist or squeeze in all the ways that conventional rigid robots cannot, according to UCSB researchers. And while it is easy to envision a world in which humans and robots collaborate — in close proximity — in many realms, emerging soft robots may help to ensure that this can be done safely, and in a way that syncs to human environments or even interfaces with humans themselves.

A small, soft actuator made of liquid metals and flexible polymers is the soft analog of an electromagnetic motor. Source: UCSB

“Some of the advantages of soft robotic systems are that they can easily adapt to unstructured environments, or to irregular or soft surfaces, such as the human body,” noted UC Santa Barbara electrical and computer engineering professor Yon Visell.

And despite their promise so far most soft robots move slowly and clumsily when compared with many conventional robots. The gap is narrowing, though, thanks to new developments in the fundamental unit of robotic motion: the actuator. Responsible for the mechanical movement of a mechanism or a machine, actuators do their work in various ways, relying on electromagnetic, piezoelectric, pneumatic or other forces.

Visell and his team of UCSB researchers have married the electromagnetic drives used in most conventional robotic systems with soft materials, in order to achieve both speed and softness.

“An interesting biological analog to the actuator described in our new work might be a fast twitch muscle,” said Visell, who along with UCSB chemistry and biochemistry professor Thuc-Quyen Nguyen, and postdocs Thanh Nho Do and Hung Phan, authored the paper “Soft Electromagnetic Actuators for Robotic Applications.”

The main challenge for Visell and the team was to build an actuator that could achieve speeds greater than what has typically been possible with soft robotic actuators, many of which depend on slow processes, such as air flow or thermal effects.

In the project, they wanted to see how far they could push the idea of having very fast, low-voltage actuation within a fully soft robotic paradigm. They based their work on the electromagnetic motor, a common type of fast and low-voltage actuator that is used in everything from electric cars to appliances, but has seen little effective application in soft robotic systems.

The team’s work has resulted in a type of actuator that is fast, low voltage and soft — and also remarkably small, just a few millimeters in size. Using unique, liquid-metal alloy conductors encased in hollow polymer fibers and magnetized polymer composites, the researchers created patterned, three-dimensional components that form the basis of soft analogs of standard electrical motors. The fibers themselves are polymer composites that the team engineered to have high thermal conductivity, greatly improving their performance.

They realized components that are each soft and stretchable, and combined them to create motor-like structures that can move things. To demonstrate, they created a tiny, millimeters-wide gripper that can close in just milliseconds, and a soft tactile stimulator that can operate at frequencies of hundreds of cycles per second.

The researchers believe these devices could find use in emerging areas such as haptics, where touch feedback is sought for applications including virtual reality and, of course haptics’ close relative — soft robotics. These soft electromagnetic actuators can be used to create tactile displays that conform to human skin, or miniature robotic tools for surgical endoscopy or other medical applications, and they said they look forward to applying these new soft robotic technologies in areas ranging from virtual reality, augmented reality, wearable technologies, healthcare and medicine.

Machine learning algorithm simulates water
Water, according to researchers at University of California at San Diego, in its molecular form is about as simple as it gets: two atoms of hydrogen and one of oxygen, joined by a chemical bond, and it is a given that this seemingly simple molecule may be the most important substance on Earth, vital for life and critical for the planet’s geology and climate. As testimony to its value, scientists have tried for years to create the most accurate molecular model of water, using increasingly powerful computers to simulate its structure in all forms, from the smallest droplets of liquid to its solidified structure of ice. Nevertheless, many questions about some of the anomalous properties of water have remained challenging to answer.

Until now, that is. A team led by researchers at UC San Diego’s Department of Chemistry and Biochemistry and the San Diego Supercomputer Center (SDSC), has used machine learning techniques to develop models for simulations of water with ‘unprecedented accuracy.’

Their work, published online in April in The Journal of Chemical Physics, demonstrates how popular machine learning techniques can be used to construct predictive molecular models, in this case of water but applicable also to other “generic” molecules, based on quantum mechanical reference data. Molecular simulations using modern high-performance computing systems are key to the rational design of novel materials with applications ranging from fuel cells to water purification systems, atmospheric climate models and computational drug design.

“Although computer simulations have become a powerful tool for the modeling of water and for molecular sciences in general, they are still limited by a tradeoff between the accuracy of the molecular models and the associated computational cost,” said Francesco Paesani, professor of chemistry and biochemistry at UC San Diego and the study’s principal investigator. “Now that we’ve proved this concept with a model of water using machine learning techniques, we are currently extending this novel approach to generic molecules, meaning that scientists will be able to predict the properties of molecules and materials with unprecedented accuracy.”

The new study builds on the highly accurate and successful “MB-pol many-body potential” for water developed in Paesani’s lab, which recently has emerged as an accurate molecular model for water simulations from the gas to liquid to solid phases. “This is a new methodology that could revolutionize computational chemistry,” said SDSC Director Michael Norman.

As reported in the paper, the researchers investigated the performance of three machine learning techniques – permutationally invariant polynomials, neural networks, and Gaussian approximation potentials – in representing many-body interactions in water. Machine learning typically involves ‘training’ a computer or robot on millions of actions so that the computer learns how to derive insight and meaning from the data as time advances.

In the quantum world, all three methods have been consistently equivalent in reproducing large datasets involving the interaction of multiple particles – “many body” phenomena such as two-body and three-body energies – as well as water cluster interaction energies, all with great accuracy, the team added.