XSIM: A structural similarity index measure optimized for MRI QSM.

Area of Science:

• Medical Imaging
• Image Processing
• Quantitative Susceptibility Mapping (QSM)

Background:

• Structural Similarity Index Measure (SSIM) is widely used for QSM quality assessment, aiming for human-like perception.
• However, SSIM exhibits bias in diamagnetic and paramagnetic tissues and can be affected by artifacts, leading to inaccurate scores.
• These limitations necessitate a more robust metric for reliable QSM evaluation.

Purpose of the Study:

• To introduce XSIM, a novel metric designed to overcome the limitations of SSIM in QSM.
• XSIM is implemented within the native QSM range with optimized parameters to address SSIM's biasing and artifact-related issues.
• The study aims to validate XSIM's superiority over SSIM and RMSE for QSM quality assessment.

Main Methods:

• Forward simulations from a ground-truth brain susceptibility map (2016 QSM Reconstruction Challenge) were used to assess SSIM, XSIM, and RMSE.
• These metrics were employed to optimize QSM reconstructions for in vivo data and a QSM abdominal phantom.
• Reconstructions from the 2019 QSM Reconstruction Challenge 2.0 were analyzed to validate XSIM across diverse algorithms.

Main Results:

• Experiments confirmed that SSIM is susceptible to biasing and artifact-induced hacking in QSM.
• XSIM demonstrated robustness against these effects, effectively penalizing streaking artifacts and reconstruction errors.
• Optimization of QSM reconstruction using XSIM resulted in less over-regularization compared to SSIM and RMSE.

Conclusions:

• XSIM is recommended as a superior metric for evaluating QSM reconstructions against ground truth.
• It effectively mitigates biasing and hacking issues inherent in traditional SSIM.
• XSIM offers a wider dynamic range of scores, enhancing its utility in QSM quality assessment.