Background: Fluorescence diagnostics uses the ability of tissues to fluoresce after exposition to a specific wavelength of
light. The change in fluorescence between normal and progression to cancer allows to see early cancer and precancerous
lesions often missed by white light.
Aim: To improve by computer image processing the sensitivity of fluorescence images obtained during examination of
skin, oral cavity, vulva and cervix lesions, during endoscopy, cystoscopy and bronchoscopy using Xillix ONCOLIFE.
Methods: Function of image f(x,y):R2 → R3 was transformed from original color space RGB to space in which vector of
46 values refers to every point labeled by defined xy-coordinates- f(x,y):R2 → R46. By means of Fisher discriminator
vector of attributes of concrete point analalyzed in the image was reduced according to two defined classes defined as
pathologic areas (foreground) and healthy areas (background). As a result the highest four fisher's coefficients allowing
the greatest separation between points of pathologic (foreground) and healthy (background) areas were chosen. In this
way new function f(x,y):R2 → R4 was created in which point x,y corresponds with vector Y, H, a*, c2.
In the second step using Gaussian Mixtures and Expectation-Maximisation appropriate classificator was constructed.
This classificator enables determination of probability that the selected pixel of analyzed image is a pathologically
changed point (foreground) or healthy one (background). Obtained map of probability distribution was presented by
means of pseudocolors.
Results: Image processing techniques improve the sensitivity, quality and sharpness of original fluorescence images.
Conclusion: Computer image processing enables better visualization of suspected areas examined by means of