Group Analysis of Functional Imaging Data using Penalized Maximum Likelihood

The value of the penalized maximum likelihood method was recently shown in assessing the test-retest reliability of functional activation (Maitra et al: MRM 2002; 48:6270). We extend this methodology to the analysis of grouped functional magnetic resonance imaging (fMRI) data. Specifically we have applied this technique to two functional paradigms, (a) pain paradigm that used a mechanical probe to provide noxious stimuli to the dorsum of the left foot, and (b) four levels of graded peripheral neuromuscular electrical stimulation (NMES). Reliability of activation maps were generated for both the paradigms. Receiver operator characteristic (ROC) curves were generated in the case of the graded NMES paradigm for each level of stimulation. These curves revealed an increase in the specificity of activation with increasing stimulus levels. Further a methodology was developed using the maximum likelihood method to determine whether the grouped reliability maps obtained from one stimulus level were significantly different from the previous levels. Our results show a significant difference (p<0.01) in reliability of activation from one stimulation level to the next. These results are in agreement with the results obtained using voxel-by-voxel measures of functional MRI signal intensities and spatial extent of activation. Besides providing information on the performance of the paradigm in a group, this methodology can also be used to optimize novel paradigms with a goal of minimizing the false detection rate.