Abstract
Purpose The term clinical functional magnetic resonance imaging (fMRI) describes an examination
with direct clinical impact on the patient. Interpretation of clinical fMRI especially
in children, however, is often difficult due to suboptimal data quality. The current
gold standard is standardized visual evaluation. To evaluate such data in an automated
and objective way, we developed an approach to identify successful sessions.
Methods Average activation inside a predefined, task-specific region of interest (ROI) is
compared with average activation in the rest of the brain, and their ratio (classification
factor [F
c]) is determined for different statistical thresholds (T). The approach was tested and validated using 239 clinical pediatric fMRI sessions
(sensorimotor, perceptive /productive language). Performance was assessed in terms
of sensitivity, specificity, and positive likelihood ratio.
Results Best performance was found for F
c ≥ 2 and T ≥ 2.5, achieving a sensitivity of 0.87 and specificity of 0.94. Comparing the different
domains, sensitivity was lowest for language production tasks, mainly due to atypical
activation foci.
Conclusion We demonstrate that an objective, automated framework for the classification of clinical
pediatric fMRI sessions may provide important additional information, supporting visual
evaluation, especially from sensorimotor and language perception domains. In the current
form, atypical or strong network activation is not easily captured.
Keywords
pediatric functional MRI - clinical fMRI - objective analysis
