Improved quantification of T2* relaxation in magnetic resonance imaging

Methods The new method has been validated in digital phantoms with actual echo times and noise levels taken from 5 patients. To achieve numerical stability at low T2* values, a novel hybrid curve fitting method was developed. For low values the residuals were weighted with the signal intensity and for high values standard exponential fitting was used. The transition between high and low regions was found by looking at the ratio between the mean and standard deviation of samples with TE >20 ms. In a pure Rician noise distribution this ratio is 1.3, ratios below this were therefore taken as indication of low T2* values. A curve fitting error map that shows the mean percentage curve fitting error for each sample was also calculated. The algorithm was implemented in the freely available cardiac image analysis software Segment (http://segment.heiberg.se). The method was validated on computer phantoms (n=140 000) with different noise levels. True T2* values were in the range of 1-100 ms, 12 echoes with TE ranges of 144 ms. Statistics were derived for the relative error in the T2* quantification as a function of the curve fitting error.