Self-Supervised Cross-Task Mutual Learning for CT Image Reconstruction
A self-supervised deep-learning framework (SS-CTML) for X-ray CT image reconstruction that needs no paired training data. From one full-view sinogram it derives sparse-view and limited-view sinograms, yielding three reconstruction tasks (full-view, sparse-view, limited-view), each handled by its own neural network. Because all three networks target the same underlying high-quality image, they are trained against cross-task mutual-learning objectives that let them supervise each other without ground-truth labels, enabling low-dose CT reconstruction that avoids the paired-data requirement of supervised methods.
SS-CTML: Self-Supervised Cross-Task Mutual Learning for CT Image Reconstruction
This paper proposes SS-CTML, a self-supervised cross-task mutual learning framework for X-ray CT image reconstruction that removes the need for paired (low-quality/high-quality) training datasets, wh…