Caltech scientists have developed a method that detects tiny, imperceptible movements at the surface of objects to reveal details about what lies beneath. By analyzing the physics of waves traveling across the surface of an object-whether that be a manufactured product or the human body-the new technique can determine both the stiffness and thickness of the underlying material or tissue. This lays the groundwork for the project's ultimate goal of enabling inexpensive, at-home health monitoring using little more than a smartphone camera.
"There is information scattered all around us in plain sight that we just haven't learned to tap into. Our work is trying to leverage that information to recover material properties from inside objects by studying tiny movements on the surface," says Katie L. Bouman , professor of computing and mathematical sciences, electrical engineering, and astronomy at Caltech and both a Rosenberg Scholar and a Heritage Medical Research Institute (HMRI) Investigator.
Bouman and her colleagues from Caltech presented the technique, called visual surface wave elastography, and its medical applications in a paper presented at the International Conference on Computer Vision in Honolulu last fall. The lead authors are Alexander C. Ogren (PhD '24) and Berthy T. Feng (PhD '25), who completed the work while at Caltech.
The group previously demonstrated that vibrations captured by a camera can be used to infer varying material properties within a 3D object of known geometry. The motivation for that work was to show that the process could be useful for nondestructive testing, such as detecting internal cracks or verifying the structural integrity of manufactured components.
The team has since turned to biomedical applications, where similar surface waves can provide insight into subsurface structure without relying on a pre-existing model of an object's geometry.
"We asked, 'Can you infer human tissue properties by the way that motion happens on the skin?' And the answer is yes," says Feng, who is now a postdoctoral fellow at MIT. Visual surface wave elastography allows scientists to measure the stiffness of underlying tissue as well as its thickness-how far the soft tissue extends before reaching bone. A change in tissue stiffness could be a biomarker for disease such as the growth of a tumor or liver disease. Meanwhile, measurements of thickness could be valuable for monitoring muscle degeneration in diseases that cause atrophy, for example.
"Because we all have cameras in our pockets, we can take frequent, inexpensive measurements of our tissue properties to track our health proactively over time," Ogren says. "We could flag concerning changes and nudge you to get it checked out. For example, a system might say, 'This region of tissue has gotten noticeably stiffer over the last month. You might want to see a doctor for a proper evaluation.'"
The new technique uses an algorithm called phase-based motion processing to detect, in video footage, the minute changes in position that occur across the skin thanks to small-amplitude waves produced by external forces such as quick pressure from a massage gun or even sound vibrations from a nearby speaker. The waves could also be excited by vibrations from a wearable device such as a watch. Importantly, the technique also quantifies the sensed movements, which are undetectable by the human eye. "Basically, the method analyzes local regions of each frame and applies standard signal processing techniques to estimate subpixel motion, resolving shifts as small as one five-hundredth of a pixel," Bouman explains.
The scientists then use spectral analysis to capture, mathematically, the propagation of those surface waves. The waves are broken down into modes that repeat periodically in time and space. The frequency describes how quickly the mode repeats in time. The wave number describes the spacing between wave peaks and troughs. By combining all of those modes, the scientists can build up a mathematical representation called a dispersion relation that represents the waves.
Importantly, the waves seen at the skin level are related to and affected by what is happening with the layers of fat, muscle, and bone beneath the surface. The scientists offer the metaphor of ocean waves coming to shore. Though we can only see the evolution of the waves and the way they break, they are affected by the rises and dips of the sea floor, for example. Similarly, the material properties and thickness of underlying soft tissue determine which pairings of wave number and frequency show up as waves on the skin.
Next, the researchers turn to a physics-based simulation of biological tissue that models a soft layer atop a much stiffer bone layer. This physics-based model allows them to identify the combination of thickness and stiffness in tissue that yields a dispersion relation that most closely matches the one derived from the video.
In the paper, the team validates its method with data from an anatomically correct simulated human leg as well as with real measurements from a gelatin model. In the case of the gelatin experiments, the new technique yielded results comparable to those of a high-precision instrument called a rheometer. And in studies of the simulated leg, the new method provided excellent estimates of thickness and stiffness at three different points in the leg despite the leg having varied nonideal geometry, just as real human bodies do.
"It is exciting to see how powerful computer vision can be in uncovering hidden properties below the surface," says Chiara Daraio , the G. Bradford Jones Professor of Mechanical Engineering and Applied Physics at Caltech and an HMRI Investigator,. "This paper shows that even in a system as complex as a human limb, the dynamic analysis of visible surface waves can reveal subsurface characteristics that are usually impossible to detect without contact."
The paper is titled "Visual Surface Wave Elastography." In addition to Bouman, Daraio, Ogren, and Feng, Jihoon Ahn (MS '25) is also an author. The work was supported by funding from the Heritage Medical Research Institute, the Department of Energy, the National Science Foundation, and the Amazon AI4Science Partnership Discovery Grant.
