The Structural Similarity Index (SSIM) has become a popular quality metric to evaluate Quantitative Susceptibility Mapping (QSM) in a way that is closer to human perception than the Root-Mean-Squared-Error (RMSE). However, SSIM may over-penalize errors in diamagnetic tissues and under-penalize them in paramagnetic tissues. Extreme susceptibility artifacts may also compress the dynamic-range, resulting in unrealistically high SSIM scores (hacking). To overcome these problems we propose XSIM: SSIM implemented in the native QSM ppm range with new susceptibility-optimized internal parameters. We validated XSIM using data from both QSM challenges. XSIM avoids bias and metric-hacking, promoting sharp susceptibility maps and preventing over-regularization.
This abstract and the presentation materials are available to members only; a login is required.