Markerless Tumor Motion Tracking in Cine Images from Megavoltage Electronic Portal Imaging Device

Lung tumor motion is challenging for image guided radiation treatment (IGRT) with a margin reduction. Locating the tumor using megavoltage (MV) cine image from electronic portal imaging device (EPID) is one solution. In this study, tumor localization was conducted for original image and processed image using deblurring and contrast enhancement. For the locating algorithm, searching of the minimum position of the mean of the sum of squared pixel differences (MSSD) between two images was used. Along with usual manual drawn masks, a novel detection mask based on Harris-Stephen (HS) corner detector was devised and tested. After conducting experiment, the detection accuracy was the best for the original set. The HS corner-based mask produced detection accuracy equivalent to or better than others. This feature-based mask creation can be used to develop automated tumor tracking or tumor registration widely needed by clinics.