Meet Duke BME’s Innovative Ultrasound Trio

The Problem with Speckle

Trahey spent his graduate career studying how to get rid of speckle, which occurs when ultrasound waves hit small structures in tissue and scatter. This scattering acts like noise and degrades an image’s quality.

Trahey continued this work after graduation when he was offered the opportunity to stay at Duke as a professor and run his own ultrasound lab. Rather than just view speckle as a problem, however, the lab probed how they could use this noise to glean further information from their images.

“When tissue moves the speckle also moves, so even though it’s noise, it’s useful noise in that regard,” says Trahey. “We did a lot of early experiments where we looked at speckle in two dimensions, and we were able track the displacement of speckle patterns and match that to track blood flow, which hadn’t been done before.”

Nearly 40 years after starting his lab at Duke, Trahey is back to exploring how to improve ultrasound images from sources of degradation, but now they are utilizing a concept known as spatial coherence.

Getting the Clearest Image Necessary

With a typical ultrasound image, a sonographer can distinguish different structures based on the differences in brightness, which indicates the strength of the ultrasound waves that are returning to the sensor. But with spatial coherence, rather than tracking the strength of the reflected waves, users track the similarity of the reflected waves recorded at slightly different locations. This process allows researchers to better distinguish borders between blood and tissue and provides valuable information about levels and sources of noise that are corrupting the ultrasound image.

Trahey’s lab has devised a way to pair spatial coherence with a technique known as adaptive imaging, which enables ultrasound systems to automatically detect and select the parameters that will provide the clearest image.

“If you’ve ever used a DSLR camera, there are so many settings you can change, like the f-stop or the shutter speed, which changes the quality of the image,” explains James Long, a PhD student in the Trahey lab who works on adaptive imaging. “Ultrasound sonographers essentially make similar changes, but they are operating under a time constraint. So while they are trying to make these changes to optimize the image quality for clinicians to read and interpret, they have to do it very quickly.”

Trahey’s new approach is really about efficiency. Rather than gathering images using standard settings and then using post-processing software to make them clearer, adaptive imaging techniques that rely on spatial coherence for image quality feedback allow researchers to quickly and easily collect the cleanest data from the get-go.

One of the applications of this approach is in fetal ultrasound imaging. Although ultrasound is considered safe, there are exposure limits set by the Food and Drug Administration. Trahey’s lab has been exploring how to use spatial coherence and adaptive imaging to gather the best image of a fetus while using the lowest level of exposure possible.

In a 2020 clinical study, Trahey and Katie Flint, a PhD student in the lab, tested 35 pregnant patients using their adaptive imaging ultrasound methods. They found that in many cases, the optimum image quality can be achieved using output levels much lower than the maximum values set by the FDA.

“We found that in fetal ultrasound, you can increase the transmit power levels to improve image quality, but after a certain point it begins to level off, where more power doesn’t lead to a clearer image,” explains Flint.

Although he didn’t intend to work in ultrasound, Trahey has helped build a large community of ultrasound engineers that extends across the country. His previous students have gone on to illustrious careers in industry, while others have found significant success in academia. This impressive list includes Muyinatu Bell, who continues ultrasound research as a faculty member at Johns Hopkins; Jeremy Dahl, an associate professor at Stanford University; Caterina Gallippi, a professor at UNC; Brett Bryam, an associate professor at Vanderbilt; Gianmarco Pinton, an associate professor at UNC; Richard Bouchard, an associate professor at MD Anderson; and Duke’s own Bill Walker, who is now the Mattson Family Director of Engineering Entrepreneurial Ventures at the Pratt School of Engineering

But Trahey has also helped cultivate an innovative ultrasound community within Duke BME, where his collaborators work in the offices next door.

Measuring Tissue Stiffness and Elasticity

As a PhD student in Trahey’s lab, Kathy Nightingale studied the potential uses of acoustic radiation force, or ARF, in diagnostic imaging. This technique allows researchers to send an ultrasonic impulse into tissue, which pushes on and momentarily displaces the tissue. At the time, Nightingale was using the technique to investigate how researchers could improve ultrasound techniques to better differentiate fluid-filled cysts from solid lesions in breast tissue.

“Generally, if you could determine that a breast mass was fluid-filled, it was a much less concerning finding than if a mass was hard and immovable,” explains Nightingale. “For my thesis, I demonstrated that you could use ARF to push on the fluid in a cyst and make the fluid swirl around. This movement could then be detected with ultrasound, making it easier to tell the difference between potentially benign and malignant masses.”

The success of this project prompted Nightingale and Trahey to explore if they could similarly use ARF to gather information about a tissue’s material properties and use it to create images. They theorized that they could use ARF to remotely palpate a target area, and then use ultrasound to image the resulting tissue motion to learn about the tissue’s stiffness.

To test this hypothesis, Nightingale and Trahey recruited the help of Mark Palmeri, who at the time was an undergraduate student at Duke.

“I became Kathy and Gregg’s research fellow, and my summer project was to simulate how ARF could deform different types of soft tissues.” says Palmeri, now a professor of the practice in Duke BME. “We published a paper in 1999 that laid out the foundation of this approach, and it was so successful that it actually led to the formation of this entirely new subfield of study in ultrasound.”

The new research avenues available with ARF led to the development of ARFI imaging, or acoustic radiation force impulse imaging, which uses the ultrasound images of tissue displacement to create a 2D map that shows relative tissue stiffness.

Nightingale, Trahey and Palmeri also began to investigate how ARF could be used to excite tissues to create shear waves, which are ripples that travel perpendicular to the acoustic disturbance. Nightingale theorized that they could use ultrasound to measure the speed of the waves as they moved across the tissue. This data could then help to specifically calculate the tissue’s stiffness. This measurement of stiffness could be used as a biomarker and help with disease diagnostics.

The approach, called SWEI, or shear wave elasticity imaging, could be paired with ARFI imaging to help researchers target and then specifically examine the elasticity of areas of tissue. Initially, Nightingale studied how the approach could be used to better characterize breast lesions as benign or malignant, but she soon recognized that breast tissue wasn’t the most optimal target if the team wanted to make an impact on diagnostic procedures.

“Breast tissue is fairly accessible, so it’s easier to do tissue biopsies without significant side effects, but the cost of a false negative is quite high,” explains Nightingale. “Unless we were almost perfect in our sensitivity with ARFI and specificity with SWEI, we were not really going to change clinical practice significantly.”

But that wasn’t true when it came to diagnosing disease in the liver.

“Liver biopsies are very painful, and they aren’t well tolerated. Physicians will therefore not perform liver biopsies frequently, so they aren’t used to monitor disease progression for things like liver fibrosis or cirrhosis, which causes scarring across the liver,” Palmeri says. “We found that we could use SWEI to numerically define the stiffness of different liver regions. This helped us identify when the tissue was diseased and scarred, allowing us to track progression in a very non-invasive way."

This liver research was particularly exciting to Siemens Healthineers, a large manufacturer of ultrasound machines. The company, which had numerous longstanding collaborations with Trahey and Nightingale, pushed to commercially translate SWEI. The technique is now a feature on most commercial ultrasound scanners.

“Over a period of eight years, this went from an idea on paper, to a simulated demonstration, to feasibility experiments, to experimental implementation, to commercial translation,” says Palmeri. “That’s the storybook fairy tale for an engineer, to have your ideas become commercially translated in a reasonable timeline."

Ultrasound Diagnostics Across the Body

Now a tenured professor in Duke BME, Nightingale is still pursuing projects that involve imaging the liver, but she’s also expanded her to work to focus on technical improvements as well as clinical projects involving muscles and organs throughout the body. On the methodology front, her team has demonstrated improved SWEI data quality when imaging with more energy in the ARF push pulses. This work is currently led by graduate student Bofeng Zhang.

One of the clinical projects, led by recent PhD graduate Cody Morris, focuses on new methods to diagnose prostate cancer. Men are regularly checked for prostate cancer as they get older. If a blood test or physical exam detects any irregularities, a physician will biopsy the prostate. Ultrasound probes are used to guide the needle directly to the prostate, but the traditional ultrasound technique isn’t sensitive enough to identify and target suspicious regions of the organ on its own.

According to Nightingale, this created problems because physicians would need to complete systematic sampling of the prostate, taking samples from different regions. And while the prostate isn’t a large organ, it’s still possible to miss cancerous cells when you’re aiming without a clear target.

In well-equipped research hospitals, a physician can perform a fusion biopsy, which involves using an MRI to identify suspicious regions and then overlaying that data with a live ultrasound to guide a targeted biopsy. This can help to ensure patients get an accurate diagnosis, but access to an MRI is inconsistent and expensive.

Nightingale, Palmeri and the rest of the team are optimistic that ARFI and SWEI could be used to help identify cancerous regions of the prostate without the expense and time-commitment of the fusion biopsy. Now they are preparing for a clinical study, led by PhD student Derek Chan, to see if their ARFI and SWEI-based ultrasound approach is as accurate as the fusion biopsy.

“We have funding to create our own ultrasound transducer that will improve our field of view, and we’ve got exciting partnerships set up with Drs. Tom Polascik, Rajan Gupta, Jiaoti Huang and Wen-Chi Foo in urology, radiology and pathology at Duke University Medical Center (DUMC),” says Nightingale. “Our vision is that this will be a good way to get more accurate diagnostic information in a single clinic visit, reducing patient stress, discomfort and the overall cost of the procedure.”

Another key effort from the lab involves using ultrasound to study muscles and muscle health. Working with Dr. Lisa Hobson-Webb at DUMC, Nightingale, graduate students Anna Knight and Courtney Trutna and research scientist Ned Rouze are investigating how they can use techniques like ARFI and SWEI to assess the progression of diseases like muscular dystrophy.

While the liver has a homogenous structure, muscle is made of fibers, and the shear waves move at different speeds depending on the orientation of the muscle. For example, these waves move faster if they are traveling in the direction of the fibers, but slow down if they travel perpendicular to them.

Ultrasound typically relies on 2D images, which create a slice of the target, and this view makes it difficult to characterize the stiffness of the muscle structure. But Nightingale and her team have developed a system that lets them study the wave propagation in three dimensions.  

“Our goal with this muscle project is to monitor and assess disease progression. We also want to see if we can track treatment response,” says Nightingale. “These things are usually done via biopsy, but you obviously can’t be sampling pieces of muscles all the time. This work is in its early stages, but ultimately we’d like to use our ARFI and SWEI systems to create a non-invasive biomarker for muscle health, just like we did for liver stiffness.”   

In addition to this work, Nightingale and her team are also coordinating with Dr. Alison Toth and Laura Pietrosimone in the departments of orthopedics and physical therapy to study how patients recover after tearing their anterior cruciate ligament, or the ACL.

In this project the team would use their new 3D scanner to assess muscle health before surgery, immediately after surgery, and then at regular intervals after to assess how the muscle is recovering and to predict how well they’ll respond to physical therapy.

“If you tear your ACL there is a long recovery, and after surgery the muscles will often atrophy. Physical therapists also don’t have a quantitative way to measure how well a patient is recovering, so it’s sometimes difficult to manage strength training and therapy,” Nightingale explains. “Being able to come up with a non-invasive way to track their recovery could be really valuable in helping patients return to an active lifestyle more efficiently.”

Bringing Artificial Intelligence to Ultrasound

Once Palmeri started ultrasound work as an undergraduate in Nightingale’s lab, he never looked back. After finishing up his undergraduate degree, Palmeri was accepted into Duke’s MD-PhD program, where he continued to work with Nightingale on SWEI research while simultaneously creating a Rolodex of future collaborators at DUMC. Now, as a professor of the practice in Duke BME, Palmeri juggles his research with his role as a leader of the department’s expanding design curriculum.

“I have this unique role in that I’m a professor of the practice, which is typically a non-research position,” he says. “But because of Duke BME’s collaborative environment, I can keep up my own arms of research, I can have a couple of students, and I can work with Kathy and Gregg to help advance projects we’re really passionate about.”

One of these passion projects centers on bringing deep learning methods to clean up ultrasound images. Rather than develop new sensors and hardware that can be added to the already-large ultrasound machines, Palmeri and MD-PhD student Ouwen Huang developed deep learning software that can automatically detect and remove speckle and other noise from ultrasound images.

“There is a push away from some of these larger, glitzy ultrasound systems and more toward portable, handheld systems, so we created algorithms that can essentially replace all those technological extras,” says Palmeri. “We named the platform MimicNet, because it was designed to allow these cheaper ultrasound systems to mimic what the commercial, high-end hospital scanners can do. It’s a great resource for a point-of-care system.”

In a separate project, Palmeri and PhD student Felix Jin are using deep learning to quantify shear wave speeds, making it easier to use SWEI for use in diagnostics. Their network can also provide feedback about the confidence of the calculation, which tells a sonographer if they need to repeat an ultrasound scan.

Now, the team is testing their program to by using it to assess if cervical stiffness can be used to predict premature delivery. In a clinical study organized by collaborators at the University of Wisconsin, Madison, the team will use the ARFI and SWEI techniques paired with the new deep learning program to test and track the stiffness of the cervix in pregnant women throughout their pregnancy.

“Cervical stiffness should decrease as a woman gets closer to her due date, but we want to see if this method can help us identify when the cervix is getting too soft too soon,” says Palmeri. “This clinical study will not only help us assess our deep learning framework, but it will also help us establish potential biomarkers that we can use to identify when a woman may be at risk for a premature delivery.”

Like Trahey before him, Palmeri’s introduction to ultrasound was more accidental than planned, but he’s still grateful that he was introduced to the field as an undergraduate student in Nightingale’s class.

“Gregg and Kathy are probably the number one reason I’ve stayed at Duke for my entire academic career. Everyone’s success is prioritized as the collective over any one person,” says Palmeri. “As the person with the most seniority, Gregg has fostered an environment where the success of everyone coming after him has been equally satisfying, and Kathy has continued that tradition. It’s really special to be a part of a community like that.”