Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) are widely used in many clinical applications. MRI plays an important role in diagnosis of disease associated with lumbar disc. MRI is limited because of its high cost and high processing time. CT scans on the other hand are much less expensive and faster. This model estimates lumbar spine MR images based on CT scan images. The lumbar spine dataset contains 641 patients, each with CT and MR image. To acquire accurate and realistic synthetic MR images by balancing qualitative and quantitative loss terms, an objective function is proposed. This function has adversarial, dual cycle-consistent, voxel-wise, gradient difference, perceptual, and structural similarity losses. A dual cycle-consistent adversarial network (DC2Anet) is proposed as a general synthesis system for semi-supervised learning. DC2Anet using semi-supervised learning can significantly perform better than supervised and unsupervised learning methods. Alternating optimization was more efficient than the integrated optimization of DC2Anet for the purpose of increase in accuracy.
Input variables : CT Scan Image
Output Variables : Estimated MRI Image
Visit Model : github.com
Additional links : mdpi.com
| Model Category | : | Public |
| Date Published | : | December, 2018 |
| Healthcare Domain | : |
Medical Technology
Provider |
| Code | : | github.com |
| Medical Imaging |
| Image Synthesis |