Multi-vendor physical intelligent network for accurate quantification of magnetic resonance spectroscopy metabolites

Motivation: The precise quantification of metabolites on MR spectroscopy using deep learning-based methods faces challenges regarding the low quantification accuracy of low-concentration metabolites. Goal(s): A new deep learning approach is proposed to accurately quantify low-concentration metabolites even in the absence of a comprehensive experimental training dataset, validated for performance across multi-vendor datasets. Approach: In this work, a method called DMnet is proposed to achieve precise quantification of low-concentration metabolites using a differentiable least squares layer network. Results: The results show that compared to state-of-the-art quantification methods, the proposed method exhibits lower MAPE for metabolite concentrations in simulated data and lower residuals in experimental data. Impact: The proposed method utilizes a differentiable least squares network layer to achieve precise quantification of low-concentration metabolites, providing more accurate quantification metrics for disease diagnosis.