Evaluation of Common Bile Duct (CBD) dilatation by CT data using synthetic MRCP data via Cycle-GAN and 3D VGG network

Abstract

MRCP (Magnetic Resonance Cholangiopancreatography) is a specialized T2-weighted MRI technique playing a crucial role in diagnosing CBD (Common Bile Duct) dilatation. In contrast, determining the exact dilatation in CT data poses considerable challenges. However, many patients prioritize the acquisition of CT images, and it is quite rare for MRCP to be taken in cases of abdominal pain. Therefore, evaluating CBD dilatation in CT data becomes essential to overcome the limitations of insufficient MRCP data in clinical settings. Previous studies have endeavored to employ deep learning algorithms for the detection and identification of CBD in MRCP images. But even when using MRCP, differentiating CBD dilatation without any segmentation label is a difficult task. In this study, we introduce a transformative solution for CBD dilatation diagnosis, enabling assessments of MRCP data without any segmented label and even with Only-CT data. The algorithm was capable of diagnosing CBD dilation without the segmented labels of MRCP. Furthermore, it could diagnose CBD dilation using only CT data by enhancing the performance through synthesis of MRCP data. This approach overcomes the limitation of insufficient data and has potential clinical applicability.