Metrics for Quantifying the Quality of MR Images

Francisco Prieto^1,2, Marcelo Guarini^1,2, Cristian Tejos^1,2, Pablo Irarrazaval^1,2

Evaluating the quality of reconstruction techniques in MRI is a difficult task. The Root Mean Square Error (RMSE) is commonly used to compare reconstructed images, although it does not correlate well with subjective perception of quality. We present two new metrics which incorporate human perception factors. One is based on the null space of the transformation from one image to the compared one, and the other on the variation of the Just Noticeable Differences map for different contrasts. Visual experiments show that both indexes provide more information than the RMSE and we propose them as metrics for comparing images.