MIT Research News' Journal
[Most Recent Entries]
[Calendar View]
Monday, August 21st, 2017
| Time |
Event |
| 12:00a |
Using machine learning to improve patient care Doctors are often deluged by signals from charts, test results, and other metrics to keep track of. It can be difficult to integrate and monitor all of these data for multiple patients while making real-time treatment decisions, especially when data is documented inconsistently across hospitals.
In a new pair of papers, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) explore ways for computers to help doctors make better medical decisions.
One team created a machine-learning approach called “ICU Intervene” that takes large amounts of intensive-care-unit (ICU) data, from vitals and labs to notes and demographics, to determine what kinds of treatments are needed for different symptoms. The system uses “deep learning” to make real-time predictions, learning from past ICU cases to make suggestions for critical care, while also explaining the reasoning behind these decisions.
“The system could potentially be an aid for doctors in the ICU, which is a high-stress, high-demand environment,” says PhD student Harini Suresh, lead author on the paper about ICU Intervene. “The goal is to leverage data from medical records to improve health care and predict actionable interventions.”
Another team developed an approach called “EHR Model Transfer” that can facilitate the application of predictive models on an electronic health record (EHR) system, despite being trained on data from a different EHR system. Specifically, using this approach the team showed that predictive models for mortality and prolonged length of stay can be trained on one EHR system and used to make predictions in another.
ICU Intervene was co-developed by Suresh, undergraduate student Nathan Hunt, postdoc Alistair Johnson, researcher Leo Anthony Celi, MIT Professor Peter Szolovits, and PhD student Marzyeh Ghassemi. It was presented this month at the Machine Learning for Healthcare Conference in Boston.
EHR Model Transfer was co-developed by lead authors Jen Gong and Tristan Naumann, both PhD students at CSAIL, as well as Szolovits and John Guttag, who is the Dugald C. Jackson Professor in Electrical Engineering. It was presented at the ACM’s Special Interest Group on Knowledge Discovery and Data Mining in Halifax, Canada.
Both models were trained using data from the critical care database MIMIC, which includes de-identified data from roughly 40,000 critical care patients and was developed by the MIT Lab for Computational Physiology.
ICU Intervene
Integrated ICU data is vital to automating the process of predicting patients’ health outcomes.
“Much of the previous work in clinical decision-making has focused on outcomes such as mortality (likelihood of death), while this work predicts actionable treatments,” Suresh says. “In addition, the system is able to use a single model to predict many outcomes.”
ICU Intervene focuses on hourly prediction of five different interventions that cover a wide variety of critical care needs, such as breathing assistance, improving cardiovascular function, lowering blood pressure, and fluid therapy.
At each hour, the system extracts values from the data that represent vital signs, as well as clinical notes and other data points. All of the data are represented with values that indicate how far off a patient is from the average (to then evaluate further treatment).
Importantly, ICU Intervene can make predictions far into the future. For example, the model can predict whether a patient will need a ventilator six hours later rather than just 30 minutes or an hour later. The team also focused on providing reasoning for the model’s predictions, giving physicians more insight.
“Deep neural-network-based predictive models in medicine are often criticized for their black-box nature,” says Nigam Shah, an associate professor of medicine at Stanford University who was not involved in the paper. “However, these authors predict the start and end of medical interventions with high accuracy, and are able to demonstrate interpretability for the predictions they make.”
The team found that the system outperformed previous work in predicting interventions, and was especially good at predicting the need for vasopressors, a medication that tightens blood vessels and raises blood pressure.
In the future, the researchers will be trying to improve ICU Intervene to be able to give more individualized care and provide more advanced reasoning for decisions, such as why one patient might be able to taper off steroids, or why another might need a procedure like an endoscopy.
EHR Model Transfer
Another important consideration for leveraging ICU data is how it’s stored and what happens when that storage method gets changed. Existing machine-learning models need data to be encoded in a consistent way, so the fact that hospitals often change their EHR systems can create major problems for data analysis and prediction.
That’s where EHR Model Transfer comes in. The approach works across different versions of EHR platforms, using natural language processing to identify clinical concepts that are encoded differently across systems and then mapping them to a common set of clinical concepts (such as “blood pressure” and “heart rate”).
For example, a patient in one EHR platform could be switching hospitals and would need their data transferred to a different type of platform. EHR Model Transfer aims to ensure that the model could still predict aspects of that patient’s ICU visit, such as their likelihood of a prolonged stay or even of dying in the unit.
“Machine-learning models in health care often suffer from low external validity, and poor portability across sites,” says Shah. “The authors devise a nifty strategy for using prior knowledge in medical ontologies to derive a shared representation across two sites that allows models trained at one site to perform well at another site. I am excited to see such creative use of codified medical knowledge in improving portability of predictive models.”
With EHR Model Transfer, the team tested their model’s ability to predict two outcomes: mortality and the need for a prolonged stay. They trained it on one EHR platform and then tested its predictions on a different platform. EHR Model Transfer was found to outperform baseline approaches and demonstrated better transfer of predictive models across EHR versions compared to using EHR-specific events alone.
In the future, the EHR Model Transfer team plans to evaluate the system on data and EHR systems from other hospitals and care settings.
Both papers were supported, in part, by the Intel Science and Technology Center for Big Data and the National Library of Medicine. The paper detailing EHR Model Transfer was additionally supported by the National Science Foundation and Quanta Computer, Inc. | | 11:00a |
Our hairy insides Our bodies are lined on the inside with soft, microscopic carpets of hair, from the grassy extensions on our tastebuds, to fuzzy beds of microvilli in our stomachs, to superfine protein strands throughout our blood vessels. These hairy projections, anchored to soft surfaces, bend and twist with the currents of the fluids they’re immersed in.
Now engineers at MIT have found a way to predict how such tiny, soft beds of hair will bend in response to fluid flow. Through experiments and mathematical modeling, they found that, not surprisingly, stiff hairs tend to stay upright in a fluid flow, while more elastic, drooping hairs yield easily to a current.
There is, however, a sweet spot in which hairs, bent at just the right angle, with an elasticity neither too soft nor rigid, can affect the fluid flowing through them. The researchers found that such angled hairs straighten when fluid is flowing against them. In this configuration, the hairs can slow a fluid flow, like a temporarily raised grate.
The results, published this week in the journal Nature Physics, may help illuminate the role of hairy surfaces in the body. For instance, the researchers posit that angled hairs in blood vessels and the intestines may bend to protect surrounding tissues from excess fluid flows.
The findings may also help engineers design new microfluidic devices such as hydraulic valves and diodes — small chips that direct the flow of fluid through various channels, via patterns of tiny, angled hairs.
“At very small scales, it’s very hard to design things with functionalities that you can switch,” says Anette (Peko) Hosoi, the Neil and Jane Pappalardo Professor of Mechanical Engineering and associate dean of the School of Engineering. “These angled hairs can be used to make a fluid diode that switches from high resistance to low when fluid flows in one direction versus another.”
Hosoi is a co-author on the paper, along with lead author and MIT postdoc José Alvarado, former graduate student Jean Comtet, and Emmanuel de Langre, a professor in the Department of Mechanics at École Polytechnique.
From cat fur to hairbrushes
“There’s been a lot of work done at the large scale, studying fluids like wind flowing past a field of grass or wheat, and how bending or changing the shape of an object affects impedance, or fluid flow,” Alvarado says. “But there’s been very little work at small scales that can be applicable to biological hairs.”
To investigate the behavior of very small hairs in response to flowing fluid, the team fabricated soft beds of hair by laser-cutting tiny holes in sheets of acrylic, then filled the holes with liquid polymer. Once solidified, the researchers removed the polymer hair beds from the acrylic molds.
In this way, the team fabricated multiple beds of hair, each about the size of a small Post-it note. For each bed, the researchers altered the density, angle, and elasticity of the hairs.
“The densest ones are comparable to short-hair cat fur, and the lowest are something like metal hairbrushes,” Alvarado says.
The team then studied the way hairs responded to flowing fluid, by placing each bed in a rheometer — an instrument consisting of one cylinder within another. Scientists typically fill the space between cylinders with a liquid, then rotate the inner cylinder and measure the torque generated when the liquid drags the outer cylinder along. Scientists can then use this measured torque to calculate the liquid’s viscosity.
For their experiments, Alvarado and Hosoi lined the rheometer’s inner cylinder with each hair bed and filled the space between cylinders with a viscous, honey-like oil. The team then measured the torque generated, as well as how fast the inner cylinder was spinning. From these measurements, the team calculated the impedance, or resistance to flow, created by the hairs.
“What is surprising is what happened with angled hairs,” Alvarado says. “We saw a difference in impedance depending on if fluid was flowing with or against the grain. Basically, hairs were changing shape, and changing the flow around them.”
“Interesting physics”
To study this further, the team, led by Comtet, developed a mathematical model to characterize the behavior of soft hair beds in the presence of a flowing fluid. The researchers worked out a formula that takes into account variables such as the velocity of a fluid and the dimensions of the hair, to calculate rescaled velocity — a parameter that describes the velocity of a fluid versus the elasticity of an object within that fluid.
They found that if the rescaled velocity is too low, hairs are relatively resistant to flow and bend only slightly in response. If the rescaled velocity is too high, hairs are easily bent or deformed in fluid flow. But right in between, as Alvarado says, “interesting physics start to happen.”
In this regime, a hair with a certain angle or elasticity exhibits an “asymmetric drag response” and will only straighten out if the fluid is flowing against the grain, slowing the fluid down. A fluid flowing from almost any other direction will leave the angled hairs — and the fluid’s velocity — unperturbed.
This new model, Alvarado says, can help engineers design microfluidic devices, lined with angled hairs, that passively direct the flow of fluids across a chip.
Hosoi says that microfluidic devices such as hydraulic diodes are one essential piece to developing complex hydraulic systems that can ultimately do real work.
“Computers and cellphones were made possible because of the invention of cheap, solid-state, small-scale electronics,” Hosoi says. “On hydraulic systems, we have not seen that kind of revolution because all the components are complex in themselves. If you can make small, cheap fluid pumps, diodes, valves, and resistors, then you should be able to unleash the same complexity we see in electronic systems, in hydraulic systems. Now the solid-state hydraulic diode’s been figured out.”
This research is supported, in part, by the Defense Advanced Research Projects Agency and the U.S. Army Research Office. | | 3:00p |
How cytoplasm “feels” to a cell’s components Under a microscope, a cell’s cytoplasm can resemble a tiny underwater version of New York’s Times Square: Thousands of proteins swarm through a cytoplasm’s watery environment, coming together and breaking apart like a cytoskeletal flash mob.
Organelles such as mitochondria and lysosomes must traverse this crowded, ever-changing cytoplasmic space to deliver materials to various parts of a cell.
Now engineers at MIT have found that these organelles and other intracellular components may experience the surrounding cytoplasm as very different environments as they travel. For instance, a cell’s nucleus may “feel” the cytoplasm as a fluid, honey-like material, while mitochondria may experience it more like toothpaste.
The team, led by Ming Guo, the Brit and Alex d'Arbeloff Career Development Assistant Professor in MIT’s Department of Mechanical Engineering, found that an organelle feels a certain resistance in cytoplasm, depending on that organelle’s size and the speed at which it moves through a cell. In particular, these characteristics determine how easily it can push against a cytoplasm’s surrounding water and move through its ever-changing web of cytoskeletal protein structures.
Certain organelles may have to work harder to make their way through cytoplasm, and may therefore feel more resistance. The researchers found that the resistance that any major organelle may feel ranges from that of a viscous fluid to an elastic, rubbery solid.
Guo and his colleagues have drawn up a phase diagram to describe the type of material that a cytoplasm would resemble, from the perspective of an organelle, given the organelle’s size and speed.
“Our main goal was to provide the most fundamental understanding of living cells as a material,” Guo says. “With this phase diagram, as long as you tell me the size and speed at which an organelle moves, I can tell you what mechanical environment it sees.”
The results, published this week in the Proceedings of the National Academy of Sciences, may help guide pharmaceutical designs. For instance, with the team’s phase diagram, scientists can tailor a drug’s size to enable it to travel within a cell with a certain amount of ease.
“A drug with a 100-nanometer diameter will feel a very different resistance than something that is 500 nanometers wide,” Guo says. “This can be a guide to understanding how a drug is delivered and transported inside a cell.”
The study’s lead author is Jiliang Hu, a former visiting student at MIT, who will join Guo’s lab as a graduate student this fall. Other co-authors include Yulong Han, a postdoc in Guo’s lab; and Alan Grodzinsky, professor of biological engineering, electrical engineering and computer science, and mechanical engineering at MIT; along with Somaye Jafari and Shengqiang Cai at the University of California at San Diego.
What a drag
Most scientists who study the transport of materials within a cell have focused on the drivers of that transport — namely, molecular motors, a family of biological agents that actively convert a cell’s energy into mechanical work to move cargo across a cell.
“But as mechanical engineers, we think the driving force is not the only part of this transport process, but that resistance of the surrounding material is actually equally important,” Guo says. “For example, it’s not just your own energy that determines how you move through a crowd — the mechanical resistance of the crowd itself can also affect your movement.”
In the case of living cells, Guo wondered whether the surrounding cytoplasm would have a similar crowding effect on the movement of major organelles such as mitochondria and lysosomes.
To test his hypothesis, he and his colleagues carried out experiments on living mammalian cells, into which they injected tiny plastic beads ranging in size from 0.5 to 1.5 microns — a range that covers most major organelles. They then dragged each bead across a cell using optical tweezers, a technique that employs a highly focused laser beam to physically move microscopic objects.
The researchers trapped and pulled each bead toward the cell edge at a constant speed and measured the force required to drag the bead a certain distance. They interpreted that force as the mechanical resistance of the surrounding cytoplasm.
They then assumed that a cytoplasm’s mechanical resistance stems from two main sources: poroelasticity and viscoelasticity. Poroelasticity originates from how fast cytoplasm can diffuse water out of a region. The group reasoned that the more poroelastic cytoplasm is, the more effort an object such as an organelle needs to make to push water out of its way.
Viscoelasticity, in the context of cytoplasm, is how fast its cytoskeleton, or web of proteins, changes configuration. A cell’s cytoskeleton serves as a sort of scaffold, made from thousands of proteins that are constantly assembling, disassembling, and reassembling. This dynamic network can feel like both an elastic solid and a viscous fluid. The faster a cytoskeleton rearranges itself, the more fluid-like it is. The researchers reasoned that an organelle would feel less resistance while moving through a more fluid-like, frequently changing cytoskeleton.
It’s all about perspective
Guo and his colleagues analyzed their experimental results and found that a bead’s size and speed were related to the type of resistance that it encountered as it was dragged across a cell. In general, the larger the beads, the more they met with poroelastic resistance, as large beads with greater surface area have to push against more water to move themselves through.
On the other hand, the faster a bead was dragged, the more it met with a solid-like resistance. As Guo explains it, “the faster you move, the more permanent [cytoskeletal] structures you would see and feel resistance to.”
The researchers drew up their phase diagram based on their experimental results. They then looked through the scientific literature for speed and size measurements, made by others, of actual organelles in living cells. They plotted these measurements onto the diagram and found that, given their size and speed, these organelles should experience a range of resistances within cytoplasm.
“If you ask a nucleus, they would tell you the cytoplasm is like honey, because they are really large and slow, and they don’t feel cytoskeletal structures — they only feel the viscous disassembled protein solution, and have very small resistance,” Guo says. “But mitochondria would say it’s like toothpaste, because they are smaller and faster, and are sometimes blocked by these constantly changing structures. A lysosome, which is even smaller and faster, would tell you the cytoplasm is actually Jell-O, because they are moving so fast, they are constantly bouncing off these structures and meeting with resistance, like rubber. So their views are limited by their own size and speed.”
Guo hopes scientists will use the group’s phase diagram to characterize other cellular components, to understand how they see their cytoplasmic surroundings.
“People can use other parameters to find out what section of the phase diagram different organelles should belong to,” Guo says. “This will tell you what kind of distinct material they would feel.” | | 5:00p |
Fusion heating gets a boost In the quest for fusion energy, scientists have spent decades experimenting with ways to make plasma fuel hot and dense enough to generate significant fusion power. At MIT, researchers have focused their attention on using radio-frequency (RF) heating in magnetic confinement fusion experiments like the Alcator C-Mod tokamak, which completed its final run in September 2016.
Now, using data from C-Mod experiments, fusion researchers at MIT’s Plasma Science and Fusion Center (PSFC), along with colleagues in Belgium and the UK, have created a new method of heating fusion plasmas in tokamaks. The new method has resulted in raising trace amounts of ions to megaelectronvolt (MeV) energies — an order of magnitude greater than previously achieved.
“These higher energy ranges are in the same range as activated fusion products,” PSFC research scientist John C. Wright explains. “To be able to create such energetic ions in a non-activated device — not doing a huge amount of fusion — is beneficial, because we can study how ions with energies comparable to fusion reaction products behave, how well they would be confined.”
The new approach, recently detailed in the journal Nature Physics, uses a fuel composed of three ion species: hydrogen, deuterium, and trace amounts (less than 1 percent) of helium-3. Typically, plasma used for fusion research in the laboratory would be composed of two ion species, deuterium and hydrogen or deuterium and He-3, with deuterium dominating the mixture by up to 95 percent. Researchers focus energy on the minority species, which heats up to much higher energies owing to its smaller fraction of the total density. In the new three-species scheme, all the RF energy is absorbed by just a trace amount of He-3 and the ion energy is boosted even more — to the range of activated fusion products.
Wright was inspired to pursue this research after attending a lecture in 2015 on this scenario by Yevgen Kasakov, a researcher at the Laboratory for Plasma Physics in Brussels, Belgium, and the lead author of the Nature Physics article. Wright suggested that MIT test these ideas using Alcator C-Mod, with Kasakov and his colleague Jef Ongena collaborating from Brussels.
At MIT, PSFC research scientist Stephen Wukitch helped developed the scenario and run the experiment, while Professor Miklos Porkolab contributed his expertise on RF heating. Research scientist Yijun Lin was able to measure the complex wave structure in the plasma with the PSFC’s unique phase contrast imaging (PCI) diagnostic, which was developed over the last two decades by Porkolab and his graduate students. Research scientist Ted Golfinopoulos supported the experiment by tracking the effect of MeV-range ions on measurements of plasma fluctuations.
The successful results on C-Mod provided proof of principle, enough to get scientists at the UK’s Joint European Torus (JET), Europe’s largest fusion device, interested in reproducing the results. Like JET, C-Mod operated at magnetic field strength and plasma pressure comparable to what would be needed in a future fusion-capable device. The two tokamaks also had complementary diagnostic capabilities, making it possible for C-Mod to measure the waves involved in the complex wave-particle interaction, while JET was able to directly measure the MeV-range particles.
John Wright praises the collaboration.
“The JET folks had really good energetic particle diagnostics, so they could directly measure these high energy ions and verify that they were indeed there,” he says. “The fact that we had a basic theory realized on two different devices on two continents came together to produce a strong paper.”
Porkolab suggests that the new approach could be helpful for MIT’s collaboration with the Wendelstein 7-X stellarator at the Max Planck Institute for Plasma Physics in Greifswald, Germany, where research is ongoing on one of the fundamental physics questions: How well fusion-relevant energetic ions are confined. The Nature Physics article also notes that the experiments could provide insight into the abundant flux of He-3 ions observed in solar flares. |
|