Circumferential Imaging: Enhancing Wound Care with AI Innovation

By Kevin Wang, Data Scientist, Net Health

Releasing new technology to clients is the most exciting part of my job. It is the cumulation of countless hours of collaboration with coworkers and clinicians, and the result of many iterations of ideas, all to solve a specific issue. In this blog post, I want to share the story of one such problem, our solution, and the insights we gained along the way. A demo for our solution on a model wound can be seen on the right. By the end of this post, you will understand why this was a problem worth solving, how our solution works, and for whom we created the solution.

Transforming Wound Care Documentation with Deep Learning

To give some background on Tissue Analytics, it is an industry-leading wound surveillance application. It uses state-of-the-art machine learning and computer vision algorithms to speed up and standardize wound measurement, helping healthcare practitioners manage wound patients more efficiently.

However, when a wound wraps around a patient’s limbs or torso, as is often the case for wounds on the arms or legs, it becomes a challenge for the clinician to accurately measure and document it, with or without Tissue Analytics. With Tissue Analytics, either the entire wound is not visible in a single image, or parts of the wound are warped, leading to an underestimation of wound size. Without Tissue Analytics, it is difficult to consistently determine the axes on which to make measurements, as the ruler used for measurements must be bent around the wound, sometimes multiple rulers must be used, and the ruler may be pressed against the open wound.

All this cumulates in inconsistent and sometimes inaccurate documentation. Circumferential imaging is designed to solve this challenge. It processes video footage in real time, converting curved wounds into 2D images that integrate directly with Tissue Analytics’ core functions for easy measurement and tracking.

Developed in Partnership with Clinicians

The development of our circumferential imaging was shaped by feedback from clinicians and continuously tested using clinical data. The earliest prototypes were refined using videos captured by clinicians from George Washington, Lakeland, and Northwell Hospital. The data they captured was critical to the testing and development of our algorithm. Our first prototype was created in Python and processed on Amazon Web Service (AWS). Clinicians would capture their videos, upload them to Tissue Analytics, and seconds later the result would be sent back. 

However, during an onsite with George Washington Hospital, consultation with clinicians revealed that this was still too slow to be used in a clinical setting. Furthermore, this method of processing did not give the clinician any feedback on the quality of their video capture. With this information on hand, the next iteration of circumferential imaging was written in C++ and Swift, and a new, smaller deep learning model was trained. These changes allowed for the circumferential imaging processing to be performed in real time, on device, giving the user real time visuals on their capture like a phone’s panorama app. By working directly with clinicians, we were able to create an intuitive tool that addressed this wound documentation challenge without interrupting the clinician’s workflow.

The Technology

The circumferential imaging algorithm combines deep learning models with classic computer vision techniques for panorama stitching to generate a 2D representation of circumferential wounds. A video is composed of multiple images, called frames, played consecutively one after another. We detect important parts, referred to as features, from the frames of the video. From these features we calculate the transformation between frames – how can we warp and alter the frames, so they match each other. After calculating the transformation between frames, it is possible to stitch a panorama. However, we can’t use the entire frame for an accurate representation of the wound. Parts of the wound disappear from frame to frame, replaced by the background; and in every frame, the wound is curved.

Our algorithm operates on the principle that a curve can be approximated by its tangent at every point, meaning a sufficiently small segment of a curve can be approximated as a flat line. Think of how the Earth is round, but when we look outside, as far as we can see, the Earth looks flat. That’s because we can only see a tiny part of the Earth, and thus, we can approximate that section of the Earth as flat. When applied to a circumferential wound, a sufficiently small segment of the wound can similarly be considered flat. We extract thin slices of consecutive frames to build our final stitched image.

And how is deep learning used in this algorithm? It is used to enhance the classic panorama algorithm for a specific use case, that is, wound imaging. We simply use a smaller version of Tissue Analytics’ proprietary wound segmentation model to extract the wound from each frame of the video. This allows the panorama stitching algorithm to focus its attention within the wound, making the generated wound as accurate as possible while ignoring the discontinuities of the background.

Committed to Supporting Healthcare Innovation

At Net Health, we pride ourselves on being at the forefront of healthcare innovation, continually developing tools that reflect the complex and evolving needs of the medical community. Circumferential imaging is an example of our commitment to making Tissue Analytics an invaluable platform that works for all wound care. Tissue Analytics is designed to support clinicians and highlight their expertise. By providing clinicians with reliable tools, we empower them to work more efficiently and make better-informed decisions. Tissue Analytics removes tedious steps in wound documentation, giving clinicians more time to focus on patient care.

Through collaboration, we can tackle new challenges in wound care and expand the use cases of Tissue Analytics. Our approach to innovation is grounded in partnership with clinicians, medical institutions, and healthcare organizations. We believe that the best solutions are those that are created not just for clinicians but with them.

The ABCs of VBC in Wound Care

Learn about value-based care in wound care with our award-winning insights

Check It Out