P₂T₂: A physically-primed deep-neural-network approach for robust T₂ distribution estimation from quantitative...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Biomedical Engineering
• Computational Neuroscience
• Medical Imaging Analysis

Background:

• Estimating T2 relaxation time distributions from multi-echo T2-weighted MRI data provides crucial biomarkers for pathologies like inflammation, demyelination, edema, and cartilage composition.
• Current deep neural network (DNN) methods struggle with low signal-to-noise ratio (SNR) clinical data and are sensitive to variations in acquisition parameters (e.g., echo times).
• These limitations hinder the clinical application and multi-institutional trial use of DNN-based T2 estimation.

Purpose of the Study:

• To develop a physically-primed DNN approach (P2T2) for accurate and robust T2 distribution estimation from MRI data.
• To improve the performance of T2 estimation in low SNR conditions and against variations in acquisition protocols.
• To enable reliable T2 distribution analysis for large-scale, multi-institutional clinical trials.

Main Methods:

• Proposed a physically-primed DNN (P2T2) incorporating the MRI signal decay forward model into the DNN architecture.
• Evaluated P2T2 against existing DNN and classical methods using 1D and 2D numerical simulations and clinical MRI data.
• Assessed model accuracy at low SNR levels and robustness against distribution shifts in echo times.

Main Results:

• P2T2 improved accuracy for low SNR levels (SNR<80), common in clinical settings.
• Achieved approximately 35% improvement in robustness against acquisition parameter shifts compared to previous DNN models.
• Generated more detailed Myelin-Water fraction maps from human MRI data than baseline approaches.

Conclusions:

• The P2T2 model offers a reliable and precise method for estimating T2 distributions from MRI data.
• P2T2 demonstrates significant improvements in accuracy and robustness, particularly for challenging clinical data.
• The approach shows promise for advancing T2-based biomarker analysis in large-scale multi-institutional studies with heterogeneous protocols.