Introduction

   Tumors are an unfortunate side effect of being alive. The good news is that as a species, we are smart enough to identify and remove the tumor while medically treating and helping the patient. Although tumor identification is not new, the procedure is typically very invasive. During the procedure, the patient is cut open, the tumor is cut out, and the wound closed and repaired.

   Due to the danger of the medical procedure, correctly identifying the tumor and its cells is paramount to saving the patient’s life and improving their quality of life. In turn, proper tumor identification begins with imaging the tumor. Below, an image of a tumor is captured with a fiber-based camera (left), and the same image is imported into Python’s Pillow library (right). Although the images look the same, a huge advantage has been gained since the image on the right is now a 3-dimensional Python array that can be computationally manipulated to provide further insight.

   In the original image, the pink coloration of the cells comes from fluorescent emission from the tumor cells after being fed a specific mix of chemicals that reach to ultraviolet light.

   Specifically, the three dimensions of the right-most image are X, Y, and RGB value. In turn, the RGB value is a tuple (or “coordinate” in color space) comprising three values: one for each of the red, green, and blue channels. In fact, the 3D array illustrated above can be broken into a separate image for the red channel, green channel, and blue channel as shown below, respectively.

   Furthermore, the fact that the original camera image is mostly blue can be used to “sanity check” the RGB channel images. For instance, an overwhelming majority of the origina camera image is blue so it is expected that the “blue” channel will be mostly bright (or white) coloration, and that is exactly what is illustrated. Similarly, the only red tones (ex. pink) in the original photograph are found in the bottom right portion of the cell cluster, so this is the same location where the red channel is expected to have its brightest (whitest) coloration, and that also matches expectations.

   But how is splitting the original image into its constituent color channels helpful? Because it enables the channels to be mathematically combined using a technique called “spectral indexing” to provide additional information. In the illustration below, spectral channels of the original image (left) have been recombined to automatically highlight the shades of pink in the image. The resulting “pink map” (below, right) can be directly compared to the original image. In doing so, it is clear that the bright pink regions in the original image map directly to the brightest (whitest) portions of the “pink map”, validating the spectral indexing approach taken during this study.

   At this point, it is straight-forward to threshold values of the pink map image to only show pixels in the original image where pink is concentrated the most. The pink map image and its thresholded counterpart are illustrated below. The power behind this image is that it provides a detailed, pixel-by-pixel map of the regions most likely to contain clusters of tumor cells. The interpretation of the image, then, is that if a medical professional were to operate on the patient and target this particular section of the patient’s body, they could likely ignore the cells surrounding the spherical clump and concentrate on extracting the bulbous tissue and the tumor contained therein.

   At the beginning of this summary, it was stated that the bright pink hue of the tumor cells was due to ultraviolet light interacting with the tumor cells and causing them to fluoresce. This, in turn, was based on the optical excitation of molecules within the liquid that the tumor cells preferentially absorbed. It is a fair question, then, to ask why ultraviolet light is necessary to use instead of blue light. After all, blue light is defined by a very similar wavelength and is easier to work with especially when it comes to transmission through optical fibers and endoscopes.

   This question was also addressed as a result of the study. In the results illustrated below, two different rats were exposed to two different chemicals, each chemical containing molecules that are known to fluoresce under ultraviolet light. In both instances, use of ultraviolet light was shown to greatly improve detectability via spectral indexing compared to blue light. Although it was easier to use blue light in an optical fiber, the molecules simply did not fluoresce as much as the ultraviolet light, greatly hindering tumor cell identification.

Conclusion

   The ability to detect and remove tumor cells in a non-invasive manner would be a game-changer for the medical industry. A large part of achieving this goal, however, is to be able to provide accurate and reliable identification of tumor cells without hurting the surrounding healthy tissue. The work done in this study confirms the viability of optically identifying tumor cells in vivo. Although not necessarily a new idea, the concept, data, and conclusions bolster existing work in the field to better human lives and advance our medical capabilities.