SAME-ECOS

SAME-ECOS decomposes multi-exponential T2 decay signals into T2 spectra and estimates myelin water fraction by combining constraint-based resolution-limit considerations with experimental condition-oriented simulations and neural network-based fitting.

Key Features:

• Experimental condition-oriented simulations: SAME-ECOS employs experimental condition-oriented simulations to tailor spectrum decomposition to acquisition-specific signal properties.
• Constraint-based resolution limits: It integrates constraint-based resolution-limit considerations to mitigate resolution limits and noise sensitivity inherent to multi-exponential fitting.
• Neural network fitting: The method leverages neural network algorithms to refine fitting and extract T2 spectra from complex datasets.
• Comparative performance metrics: On simulated data and six in vivo brain datasets, SAME-ECOS achieved over 15% higher cosine similarity for T2 spectra and more than 10% lower mean absolute error in myelin water fraction compared with non-negative least squares (NNLS).
• Noise robustness: SAME-ECOS exhibited enhanced robustness to noise relative to NNLS.
• Distinct peak separation: In in vivo mean T2 spectra, it separated the myelin water peak from intra/extra-cellular water peaks where NNLS did not.
• Processing speed: The approach is approximately 30 times faster than NNLS, enabling whole-brain analysis within three minutes.
• Result consistency: Myelin water fraction results from SAME-ECOS showed high correlation with NNLS across study participants.

Scientific Applications:

• Neuroimaging T2 spectrum decomposition: Provides accurate decomposition of multi-exponential T2 decay signals for analysis of brain tissue properties.
• Myelin water fraction quantification: Enables estimation of myelin water fraction (MWF) in vivo with reduced mean absolute error compared to NNLS.
• Whole-brain multi-component T2 analyses: Facilitates rapid whole-brain multi-component T2 decay analyses in research studies.

Methodology:

Computational steps include experimental condition-oriented simulations, constraint-based resolution-limit considerations, calculation and simulation operations, and neural network model training.