By modeling spatial variant z-spectrums into low-dimensional subspace, we developed Implicit Regression in Subspace (IRIS), which is an unsupervised denoising algorithm utilizing the excellent property of implicit neural representation for continuous mapping. Denoising, as one of the post-processing stages for CEST data, can effectively improve the accuracy of CEST quantification.
The clinical application of CEST is constrained by its low contrast and low signal-to-noise ratio (SNR) in the acquired data.
Our unsupervised denoising algorithm (IRIS) improves the SNR and sensitivity of CEST MR imaging.
It leverages an extremely lightweight neural network while achieving high performance in representing continuous signals.
Qualitative and quantitative evaluations on both synthetic and in-vivo datasets demonstrate the outperformance of IRIS over other methods in terms of noise reduction while maintaining the integrity of the underlying CEST signal.
