ROI based Multi-parameter Quantitative Network(RMQ-Net) with Uncertainty-awareness for Quantitative UTE MRI Study of Cartilage

Motivation: Quantitative MRI (qMRI) studies of cartilage regions need both regional segmentation and pixel-wise fitting analysis, which can be time-consuming and subject to inter-individual variability. Goal(s): To design a deep neural network for simultaneous qMRI mapping and accurate tissue segmentation. Approach: By leveraging different scan sequences, we proposed a RMQ-net with Uncertainty-awareness(UA) module, or UA-QMR-net. A majority-voting strategy was applied for robust cartilage segmentation and accelerated qMRI analysis. Results: The results demonstrated that the UA-RMQ-net achieved higher performance than the original RMQ-net for both UTE-T1 and UTE-T1r analyses of articular cartilage. Impact: By leveraging information from different scan sequences, the proposed UA-RMQ-net could obtain higher performance for accelerated qMRI analysis.