A Temporal-compensated Structure-preserving Enhancement Network (Tco-SEN) for Abdominal Four-dimensional Magnetic Resonance Imaging

Motivation: Four-dimensional Magnetic Resonance Imaging (4D-MRI) shows promise for motion management in abdominal radiotherapy. However, the prevalent undersampling often hampers its image quality. Goal(s): To enhance the image quality of 4D-MRI, we propose Tco-SEN, a deep-learning model to exploit its properties. Approach: Tco-SEN employs a two-stage architecture and a customized loss penalty, enabling effective restoration of detailed features and preservation of anatomical structures. Results: Compared to state-of-the-art algorithms, Tco-SEN significantly enhances image quality by improving spatial resolution, reducing motion artifacts and noise, and preserving delicate structures. Furthermore, our method enhances the accuracy of subsequent motion modeling in 4D-MRI, highlighting its potential for clinical applications. Impact: Tco-SEN effectively improves the image quality of 4D-MRI, benefiting more accurate tumor delineation and motion estimation. This advancement promotes the application of 4D-MRI in cancer radiotherapy, ultimately enhancing the accuracy of abdominal cancer radiation treatment.