P.298Will qNMRI-based FF trajectories help in the prediction of disease progression in Duchenne muscular dystrophy: a study in forearm muscle?

Abstract

Several studies in Duchenne muscular dystrophy (DMD) have shown trajectories of mostly functional outcome measures across several years. Additionally, longitudinal quantitative nuclear magnetic resonance imaging (qNMRI) data have been subject of a series of studies in recent years. Here, we present longitudinal qNMRI data in forearm for up to seven years in the AFM Genethon-sponsored DMD natural history study. We wanted to investigate how well disease progression could be predicted using a sigmoidal model. A 3-point Dixon-based fat-water separation NMRI sequence (TE1/TE2/TE3= 2.75/3.95/5.15 ms; TR = 10 ms) was obtained on a clinical 3T Siemens scanner at one-year intervals in forearm muscles. Thirty-five DMD boys (12.6±3.3 years) were included of which 29 subjects had at least two NMR scans. A two-parameter sigmoidal model was fitted on each FF trajectory with the following constraints: FF at age=0 years is smaller than 1%; slopes > 0; asymptote at 90% FF. Trajectories were fitted in four different ways: using 1, 2, 3 and 4 time points (with 1 time point being every next year) and, consequently, the difference between the actual measured FF and the fitted FF value on subsequent time points was calculated. This initial analysis was performed on 8 of 35 subjects who all had 5 consecutive yearly NMR exams. When using just the first time point to fit the trajectories an average error on the actual measured FF of 4.0%, 5.7%, 5.9% and 8.4% were observed for the 2nd, 3rd, 4th and 5th time point, respectively. Using the first two time points in the fit, an error of 4.3%, 8.4% and 11.0% were observed for the 3rd, 4th and 5th time point, respectively. Using the first three time points in the fit, an error of 4.9% and 7.0% were observed for the 4th and 5th time point, respectively. Using the first four time points, an error of 3.8% was obtained for the 5th time point. These data indicate that, at least when using a 3rd and 4th time point in the sigmoidal fitting, the error on the FF values for subsequent time points decreases. In this subset, adding a 2nd time point did not decrease the difference with the actual measured FF values, due to some trajectories deviating strongly from the imposed sigmoid. Further investigations on the models used are necessary to assess whether FF trajectories can actually be used to predict disease progression.