Joint spectral quantification of MR spectroscopic imaging using linear tangent space alignment-based manifold...

Area of Science:

• Medical Imaging
• Computational Biology
• Data Science

Background:

• Magnetic Resonance Spectroscopic Imaging (MRSI) provides valuable metabolic information but faces challenges in accurate spectral quantification.
• Existing methods like QUEST and subspace-based approaches have limitations in handling the complexity and noise of MRSI signals.

Purpose of the Study:

• To develop and validate a novel manifold learning-based method for joint spectral quantification in MRSI.
• To leverage the intrinsic low-dimensional structure of MRSI signals for improved metabolite concentration estimation.

Main Methods:

• A Linear Tangent Space Alignment (LTSA) model was employed to represent MRSI signals, aligning local metabolite subspaces to a global low-dimensional manifold.
• Basis functions were predetermined via quantum mechanics simulations, and model parameters were estimated using noisy MRSI data with spatial smoothness and sparsity constraints.

Main Results:

• The proposed LTSA method demonstrated superior performance compared to QUEST and subspace-based methods in both numerical simulations and in vivo human MRSI data.
• Qualitative improvements included reduced noise and artifacts in metabolite concentration maps.
• Quantitative analysis showed higher spectral quantification accuracy, as measured by normalized root mean square errors.

Conclusions:

• Linear Tangent Space Alignment-based manifold learning offers a significant advancement for joint spectral quantification in MRSI.
• The method enhances the accuracy and reliability of metabolite concentration measurements from MRSI data.