RESUMO
RATIONALE AND OBJECTIVES: Registration is an important process to detect interval changes between two chest x-ray images. However, the conventional registration methods suffer from misregistration because of the difference in rotation angles of human body around an axis parallel to the x-ray films, such as anteroposterior inclination. Such difference causes permutation of the shadows between the two images, which makes registration difficult. This article proposes a novel registration method in cases where two chest x-ray images have different rotation angles. MATERIALS AND METHODS: Twelve x-ray images taken from a chest phantom and four chest photofluorograms of two patients were used to evaluate the performance. First, the proposed algorithm estimates the rotation angles of the body from the pair of two x-ray images based on the function describing the relationship between a point in the current image and that in the previous image, which is derived from a three-dimensional rotational model of the body. Then it aligns two images according to the function. RESULTS: From the results of estimating rotation angles, it was found that proposed method can estimate the angles with an error of less than 1 degrees. Then two physicians evaluated the subtraction images and confirmed that this approach makes it possible to detect the interval changes accurately even if there are permutations of shadows in the x-ray image. CONCLUSIONS: The proposed method is superior to the conventional one when two chest x-ray images have different rotation angles.