Synthesizing CT image from unpaired MR images in the pelvic area

Abstract

Computed tomography (CT) is one of the prevailing medical imaging modality since it was introduced. Due to cost-effectiveness and fast scanning time compared to magnetic resonance imaging (MRI), CT has been widely used in numerous medical practices such as detection of tumors, hemorrhage and diagnosis of abdominal diseases. Also, since CT directly provides the electron density information, CT is used to calculate dose for radiotherapy and to perform attenuation correction of SPECT and PET images. However, exposure to CT scans can increase the chance of getting cancer and, using radiocontrast agents to enhance the contrast between tissues may cause reactions such as mild, skin rashes and nephropathy. Along with CT, MRI is one of the widely used imaging tools in the medical field. Depending on the sequence, MRI can provide anatomical information and functional information without any radiation exposure. The tissue contrast of MR images is superior to CT images and, this sometimes makes MRI to be taken with a combination of other imaging methods, like CT, for accurate localization and segmentation. Besides, since MRI provides superior tissue contrasts to CT, additional MRI enables the detection of lesions that are difficult to distinguish. Yet, the acquisition of both imaging modalities can be a time-consuming process and might be a burden to patients. Here, we proposed a convolutional neural network (CNN) that generates synthetic CT (sCT) images from MR images, on the pelvic area. Purpose of this network is to reduce the time and burden of patients, as well as radiation risk from CT imaging.