Abstract
Abstract
Objective. Quantitative magnetic resonance imaging (qMRI) can noninvasively acquire parametric maps, providing quantitative biomarkers for clinical diagnosis. Owing to their distinct physical origins, various parametric maps have inherent differences in numerical scale. The existing deep learning methods primarily focus on end-to-end feature learning in the image domain without fully considering these inherent differences, which exhibit limited generalizability when processing diverse quantitative tasks. To improve the generalizability and robustness of the deep learning model for parameter mapping tasks and to deliver dependable quantitative references for clinical diagnosis, we propose a wavelet-guided diffusion model. Approach. We extend the discrete convolution wavelet transform (DCWT) to two-dimension and combine it with a conditional denoising diffusion model to develop a diffusion-based discrete convolution wavelet transform (Diff-DCWT) for parameter mapping. Diff-DCWT achieves accurate parameter mapping under the guidance of wavelet coefficient maps, thereby reducing reliance on image domain learning. The performance of Diff-DCWT is evaluated through simulation and in vivo experiments across T2, T2 * , and ADC mapping tasks. Main Results. Both simulation and healthy volunteer experiments demonstrate that T2, T2 * , and ADC maps obtained from Diff-DCWT maintain excellent visual quality and quantitative accuracy. Clinical case studies validate the efficacy of T2 and ADC maps from Diff-DCWT for stroke staging. In glioma characterization, Diff-DCWT enables accurate T2 mapping, facilitating clear differentiation between lesional and normal brain regions. Significance. Diff-DCWT establishes a high-fidelity framework for parameter mapping, with its clinical efficacy validated in stroke staging and glioma characterization. Moreover, this work demonstrates the potential of wavelet-domain prior guidance for deep learning-based qMRI applications.