Hyperspectral imaging (HSI) devices, calibrated to new National Institute of Standards and Technology standard reference spectra, could be used as noninvasive diagnostic tools for discriminating between healthy and diseased tissue and for providing new insight on the physiology of tissue injury and healing. Many physicians would prefer the completely noninvasive HSI methods available to measure the amount of oxygen in tissue — a key factor in wound healing; at present, this process requires biopsies or transcutaneous techniques limited to a single point. David Allen takes readings using a NIST standard reflectance diffuser before scanning a wound area on an anesthetized pig. Illumination comes from the broadband lights on the hyperspectral camera. (Images: NIST) “HSI isn’t being used routinely in the clinic yet,” said David Allen of NIST’s Physical Measurement Laboratory. “One big reason is that standards aren’t in place. Individual researchers doing their own experiments show positive results. But when you start comparing an instrument in one lab to an instrument in another, the data are typically inconsistent.” Factors that limit repeatable results include biological variability (variations in skin pigmentation, tissue density, lipid content and blood volume changes) as well as sensor variability related to calibration and best practices in the measurement protocol. While the biological variability is beyond researchers’ control, sensor variability can be minimized. In pursuit of lessening sensor variability, Allen and his colleagues produced the first prototype “digital tissue phantoms” derived from benchtop simulations and in vivo wound imaging. The digital tissue phantoms are a set of specific spectral signatures and images that correspond to different states of hemoglobin oxygenation due to ischemia, or inadequate blood flow. A hyperspectral line camera with four broadband light sources acquires images with the lens at center. A typical scan captures images at hundreds of discrete wavelengths that differ by as little as 2 nm. Eventually, more extensive and clinically validated versions of the phantoms will be used to calibrate spectral imaging devices of various kinds. Those devices will detect telltale spectral evidence of ischemia, revascularization, assorted pathologies and other conditions that could suggest a tissue’s viability at the wound site on a microvascular scale — even during surgery. For that to happen, though, there must be a well-characterized and clinically validated correspondence between a particular spectral signature and particular tissue conditions. Allen and scientists from the biomedical engineering department at Ohio State University in Columbus imaged carefully manipulated ischemic wounds in an anesthetized pig. “This is the first time anybody has looked at a porcine skin flap animal model hyperspectrally,” Allen said. “The collection of the hyperspectral data for use as a reference is a small but very significant milestone.” A hyperspectral image such as this one, which integrates scans from 240 different wavelengths, can be used as a digital tissue phantom. The “data cube” for this image of a 15 × 5-cm ischemic wound combines readings from 740 individual lines and about 600 rows at 12 bits per pixel. On two areas of the animal’s back, flaps of skin were raised, silicone-plastic sheets were placed beneath the skin to inhibit reperfusion, and the incisions were closed. A skin flap without a plastic sheet and an untreated control area were adjacent to the treated area. Allen scanned all the areas with the hyperspectral imager, using illumination from a standard broadband light source. A typical scan encompassed 240 different wavelengths, spaced about 2 nm apart, ranging from 400 to 880 nm. As a reflectance reference, each image also included light from a NIST-traceable standard diffuser. Readings for each wavelength were stacked into “data cubes” for each scanned position. “We found that we can reproduce the tissue spectrum — including the oxygenation level of hemoglobin — to within one standard deviation, well within the natural variability of the tissue,” Allen said. Allen’s group received NIST funding to develop standards for the field. One part of the award supported use of a device called the hyperspectral image projector (HIP), which accurately reproduces complex spectral-spatial images by precision control of digital micromirror devices. A diagram of the digital micromirror device used in the hyperspectral image projector. It employs a 1024 × 768 array of the mirrors, each of which is about 14 µm wide — about one-seventh the width of a human hair. The hyperspectral scenes projected were medically significant, so they are referred to as digital tissue phantoms. These realistic medical scenes can be produced repeatedly without the variability and expense involved in repeating a medical procedure every time that a hyperspectral image is evaluated. Allen's group has already generated HIP-projectable spectral signatures that correspond to different levels of oxygenation in tissue. Currently, accumulating the data for digital tissue phantoms is a time-consuming process: Each scan takes from tens of seconds to a minute or more. “That’s the state of the technology right now,” Allen said. “But in the future, we’ll have ‘snapshot’ hyperspectral scans for real-time imaging, including video. Eventually, we want to get to the point at which you can see the blood perfusing through the tissue.” In addition to optical medical imaging, the scientists are also investigating HSI's potential for environmental and defense applications, such as diseases of coral reefs and the detection of hidden explosive devices. For more information, visit: www.nist.gov