WEST LAFAYETTE, IN – You may already be using your smartphone for remote medical appointments. Why not use some of your vehicle’s sensors to collect medical data? That’s the idea behind the AI-driven technology developed at Purdue University. This technology can use your smartphone’s camera to detect and diagnose medical conditions, such as anemia, more quickly and accurately than highly specialized medical devices being developed for that task.
“Smartphones have at least 15 different sensors, and our goal is to leverage those sensors to give people access to healthcare outside of the doctor’s office,” says lead researcher. said Young Kim, professor and associate director of research at Purdue University, Weldon. Faculty of Biomedical Engineering. “To our knowledge, we believe we have demonstrated the fastest hemodynamic imaging in existence using a commercially available smartphone.”
Smartphone cameras are useful, but they are of limited medical use because they can only take measurements of the red, green, and blue wavelengths of light within each pixel. Hyperspectral imaging can capture all wavelengths of visible light at each pixel and can potentially be used to detect various skin and retinal conditions, as well as some cancers. Researchers are exploring applications of hyperspectral imaging in healthcare, but most of that work is aimed at improving specialized equipment that is relatively bulky, time-consuming, and expensive. By combining deep learning and statistical methods with knowledge of light-tissue interactions, researchers at Purdue University were able to reconstruct the full spectrum of visible light at each pixel in a typical smartphone camera image. A patent-pending approach from a lab with mobile health expertise could improve access to healthcare.
As reported in PNAS Nexus, the research team used a commercially available hyperspectral imaging device in collecting information on blood oxygen movements in volunteer eyelids, models mimicking human tissue, and chicken fetuses. I tested the method against The results show that smartphone cameras produced hyperspectral information as accurately, faster, and cheaper than information captured using specialized equipment. The smartphone approach can produce images in 1 millisecond that would take him 3 minutes to capture with conventional hyperspectral imaging.
Kim said the research reported at PNAS Nexus focuses on building hyperspectral imaging algorithms for smartphones rather than on specific applications. However, in other studies, the team used an approach that measures blood hemoglobin for tissue oximetry and inflammation. Kim’s lab used a computational approach the researchers called “hyperspectral learning.”
The process starts by putting your phone’s camera on a super slow-motion setting to generate a video at about 1,000 frames per second. Each pixel in each frame contains information about the intensity of red, green, and blue colors. This information is provided through machine learning algorithms that infer the full spectral information of each pixel. It is used to generate measurements of blood flow, specifically the amount of oxygenated and deoxygenated hemoglobin within each pixel. These hemodynamic parameters can also be used to create images and videos showing a subject’s oxygen saturation over time.
As with traditional machine learning, the team trains an algorithm on a dataset, feeds it smartphone images and corresponding hyperspectral images, and fine-tunes the algorithm until it predicts the correct relationship between the two datasets. However, by building algorithms using equations derived from tissue optics (an approach sometimes called “informed learning”), researchers require much smaller training data sets.
Also, while traditional hyperspectral imaging equipment has to collect large amounts of data and is limited in either spectral or temporal resolution, the team’s approach enables video images hundreds of times smaller than hyperspectral imaging files. You start with a file, so you can maintain high standards on both fronts. .
“Usually, collecting this information efficiently requires trade-offs. However, our approach can simultaneously achieve high spatial and spectral resolution,” says lead author and graduate student in the Kim lab. One Yoohyun Ji said. The lab is currently working on applying the method to other mobile health applications such as cervical colposcopy. and retinal fundus photography.
Kim disclosed his innovation to the Purdue Research Foundation Office of Technology Commercialization, which has applied for a patent to protect its intellectual property. Industry partners interested in further developing or commercializing their innovations should contact Senior Business Development Manager Patrick Finnerty at PWFinnerty@prf.org on 2019-KIM-68586.
“MHealth Hyperspectral Learning for Instantaneous Spatial Spectral Imaging of Hemodynamics” was prepared with support from the National Institutes of Health and the Ralph W. and Grace M. Showalter Trust.
Writer/Media Contact: Mary Martilay, mmartial@purdue.edu
sauce: Young Kim, youngkim@purdue.edu
