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

Hadas Ben-Atya; Moti Freiman

doi:10.1016/j.compmedimag.2023.102240

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

Comput Med Imaging Graph. 2023 Jul:107:102240. doi: 10.1016/j.compmedimag.2023.102240. Epub 2023 May 9.

Authors

Hadas Ben-Atya¹, Moti Freiman²

Affiliations

¹ Faculty of Biomedical Engineering, Technion - Israel Institute of Technology, Haifa, Israel. Electronic address: hds@campus.technion.ac.il.
² Faculty of Biomedical Engineering, Technion - Israel Institute of Technology, Haifa, Israel.

PMID: 37224742
DOI: 10.1016/j.compmedimag.2023.102240

Abstract

Estimating T₂ relaxation time distributions from multi-echo T₂-weighted MRI (T₂W) data can provide valuable biomarkers for assessing inflammation, demyelination, edema, and cartilage composition in various pathologies, including neurodegenerative disorders, osteoarthritis, and tumors. Deep neural network (DNN) based methods have been proposed to address the complex inverse problem of estimating T₂ distributions from MRI data, but they are not yet robust enough for clinical data with low Signal-to-Noise ratio (SNR) and are highly sensitive to distribution shifts such as variations in echo-times (TE) used during acquisition. Consequently, their application is hindered in clinical practice and large-scale multi-institutional trials with heterogeneous acquisition protocols. We propose a physically-primed DNN approach, called P₂T₂, that incorporates the signal decay forward model in addition to the MRI signal into the DNN architecture to improve the accuracy and robustness of T₂ distribution estimation. We evaluated our P₂T₂ model in comparison to both DNN-based methods and classical methods for T₂ distribution estimation using 1D and 2D numerical simulations along with clinical data. Our model improved the baseline model's accuracy for low SNR levels (SNR<80) which are common in the clinical setting. Further, our model achieved a ∼35% improvement in robustness against distribution shifts in the acquisition process compared to previously proposed DNN models. Finally, Our P₂T₂ model produces the most detailed Myelin-Water fraction maps compared to baseline approaches when applied to real human MRI data. Our P₂T₂ model offers a reliable and precise means of estimating T₂ distributions from MRI data and shows promise for use in large-scale multi-institutional trials with heterogeneous acquisition protocols. Our source code is available at: https://github.com/Hben-atya/P2T2-Robust-T2-estimation.git.

Keywords: Deep-learning; Microstructure imaging; Myelin-Water imaging; Physically-driven deep-learning; T(2) relaxometry.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Humans
Magnetic Resonance Imaging* / methods
Neural Networks, Computer
Osteoarthritis*
Signal-To-Noise Ratio
Software