A multimodal image feature extraction method for x-ray grating phase contrast computed tomography based on monogenic signal

Traditional computed tomography (CT) based on x-ray absorption imaging has made great progress in clinical medicine, and CT combined with x-ray phase contrast imaging (XPCI) technology has become a new research hotspot in recent years. XPCI can separate the attenuation, refraction, and scattering signals of the object and retrieve three types of feature images known as absorption contrast image, differential phase contrast image, and dark field contrast image. However, the quality of CT images is always degraded due to noise and reconstruction artifacts, which makes feature recognition methods for CT images necessary. Most of the existing CT image recognition algorithms are focused on AC-CT images, with little attention paid to other contrast images. Herein, a new method is proposed, named the variable kernel multi-scale adaptive monogenic signal phase consistency model (VK-MA PC model), which constructs monogenic signals with corresponding filters according to the characteristics of different contrast images. The model obtains better image features by using multi-scale analysis and optional pre-decomposition, which make images decomposed into different levels. Experiments on 4D extended cardiac-torso (XCAT) human body simulation data and laboratory fish XPCI-CT data demonstrate the potential applicability of the VK-MA PC model in the field of XPCI-CT.