DeepEMC-T2 mapping: Deep learning-enabled T2 mapping based on echo modulation curve modeling | JoVE Visualize

Area of Science:

Medical Imaging
Artificial Intelligence in Radiology
Quantitative MRI

Background:

Accurate T2 relaxation time quantification is crucial for Magnetic Resonance Imaging (MRI) analysis.
Traditional Echo Modulation Curve (EMC) T2 mapping requires numerous echoes and computationally intensive pixel-wise dictionary matching.
Existing methods face limitations in efficiency and speed for clinical applications.

Purpose of the Study:

To develop and evaluate DeepEMC-T2 mapping, a deep learning-based framework for efficient and accurate T2 quantification.
To reduce the number of required echoes for T2 mapping without compromising accuracy.
To overcome the limitations of standard EMC-T2 mapping, particularly the need for extensive data and complex processing.

Main Methods:

A modified U-Net deep learning architecture was employed to directly estimate T2 and proton density (PD) maps from multi-echo spin-echo (MESE) images.
The network incorporated novel features to enhance T2/PD estimation accuracy.
Extensive training and validation were performed using 67 axial MESE datasets, with generalizability tested on 57 coronal datasets acquired under varying parameters.

Main Results:

DeepEMC-T2 mapping demonstrated high accuracy, with T2 estimation errors ranging from 1% to 11% and PD errors from 0.4% to 1.5% using as few as three echoes.
The framework showed robust generalizability across different scan orientations and parameters due to joint training strategies.
Improvements in accuracy were linked to increased echo spacing and the inclusion of longer echoes, with all proposed network features contributing to better T2 estimation.

Conclusions:

DeepEMC-T2 mapping provides a simplified, efficient, and accurate method for T2 quantification directly from MESE images, eliminating the need for dictionary matching.
The ability to achieve accurate T2 estimation with fewer echoes facilitates faster imaging protocols.
This advancement allows for increased volumetric coverage and/or higher slice resolution within standard scan times, enhancing clinical utility.