Contrast Enhancement of Portal Images with Adaptive Histogram Clip

The radiotherapy is a widely used and effective method for cancer treatment where a highenergy beam is targeted at the tumor volume. To properly deliver the required radiation to the tumor and to prevent surrounding tissue from destruction, the beam has to be shaped as prescribed by the physician. This requires registration of two images: a reference image and an image obtained during the radiation delivery called portal image. However, portals are notorious with their poor quality which hampers the anatomical structure delineation and makes image comparison a challenge. Therefore, the accurate registration requires a preprocessing step aimed at the contrast enhancement. Many contrast enhancement techniques are based on the histogram equalization (HE) . The idea is to transform the distribution function of the original image and make it as close as possible to a distribution of prescribed shape. As a result some grey levels are combined into one bin and the dynamic range may increase. However, when applied to a whole image HE will ignore local peculiarities and may not perform very well. To avoid this an adaptive histogram equalization technique (AHE) was developped [1, 2]. Here HE is applied in a window sliding over the image pixel by pixel and the grey level of the central window pixel is transformed, accordingly. This technique performs better than HE but at the expense of much more computational time. To reduce it, an interpolated AHE was developed by P i z e r e t a l. [3]. In their approach the image is divided into nxny contextual regions of size dxdy and histograms are evaluated and equalized for all regions. The greys of the original image are transformed using a bilinear interpolation of the equalized histograms of every four neighbor regions. A slightly different approach was suggested in [5]. While the obtained quality is almost as good as that obtained by AHE, the processing time is significantly reduced. A major shortcoming of these approaches concerns the overenhancement of relatively homogeneous regions. Thus, the noise in the background of portal images may be drastically increased. Also, field edges may be significantly blurred. БЪЛГАРСКА АКАДЕМИЯ НА НАУКИТЕ . BULGARIAN ACADEMY OF SCIENCES