A non-linear multi-scale blending algorithm for fluorescence bladder images

Abstract

The composition of panoramic images of the internal urinary bladder wall from single endoscope images can strongly support the off-line documentation and surgery planning for cancer treatment, as well as assist the reidentification of multi-focal tumors during a cystoscopy. Unlike white light endoscopy, fluorescence techniques such as photodynamic diagnostics (PDD) lower the risk of missing flat and small tumors due to an enhanced tissue contrast. As a result of the low illumination power and the free hand movement of the endoscope, PDD video sequences usually show strong variations in illumination and resolution. During the subsequent mosaicking and blending process, these effects impede the preservation of original and unbiased input image information in the composed panoramic overview image, and make it difficult to avoid visual interpolation artifacts. Thus, a non-linear intensity based multi-scale blending method for fluorescence images is developed. Based on a highest intensity decision, a region mask modeling the endoscopic illumination characteristic is used to weight the input images on several sub-bands of a Laplacian pyramid. In comparison to basic linear interpolation and standard multiscale methods the new method preserves high fluorescence intensities as well as fine vessel structures in the final image composition. Furthermore the adaptive characteristics of the blending algorithm allow the physician to move the endoscope more freely along the bladder wall during the image mosaicking process.A. Behrens ( ) · M. Guski · T. Stehle · S. Gross · T. Aach

Stitching and Surface Reconstruction From Endoscopic Image Sequences: A Review of Applications and Methods

“…An acceleration of the method presented by Miranda-Luna et al was described by Hernandez-Mier et al in 2010 [65]. Behrens et al have investigated methods for image stitching of fluorescence cystoscopy in several publications [11]- [13], [16], [17] (2008-2013). They explored the creation of panorama images from cystoscopies with pure image-processing methods, as well as navigation support using inertial sensors.…”

“…This function is kept constant for all images, implicitly assuming an invariant vignetting effect. This approach has been followed by Behrens et al [16], Bouma et al [26], and Soper et al [133]. Mountney and Yang [108] have also addressed the vignetting problem in SLAM-based mosaicking.…”

“…The basic linear alpha blending (with or without feathering) usually produces smooth transitions but also tends to blur image structure. This effect is reduced with multi-band blending, which is therefore applied by Behrens et al, who also introduced a non-linear component to prefer brighter images over darker ones [16]. In [160] and [162], Weibel et al present a blending strategy as energy minimization problem.…”