Motivation: Quantitative MRI offers direct measurement of tissue parameters but is limited by the need for specialized sequences and lengthy acquisition times. Goal(s): This study aims to create an efficient method for estimating T1 and T2 maps from conventional MRI to facilitate wider clinical adoption. Approach: We propose a weakly-supervised learning method to estimate T1 and T2 maps from conventional weighted MRI, leveraging large amounts of unlabeled data while requiring minimal labeled data for fine-tuning. Results: The proposed method achieved high estimation accuracy with visual similarity to label maps, suggesting its potential for more efficient and practical adoption of quantitative MRI. Impact: This work enhances the practicality of quantitative MRI by reducing data requirements and improving generalizability, paving the way for broader clinical adoption and efficient retrospective mapping of T1 and T2 from conventional MRI with minimal labeled data.
Xu et al. (Tue,) studied this question.