Poster F81, Tuesday, March 27, 8:00-10:00 am, Exhibit Hall C
Optimizing fMRI experimental outcomes via neuroadaptive task designs
Ming-Hua Chung1, Bradford Martins1, G. Andrew James1, Anthony Privratsky1, Clinton D. Kilts1, Keith A. Bush1; 1University of Arkansas for Medical Sciences
Task-based fMRI is a widely-used tool for studying the neural underpinnings of cognition in both healthy and clinical populations. There has been growing interest in mapping individual differences in fMRI task behavior and neural organization, both within and between clinical samples. By utilizing a neuroadaptive task designs accounting for individual differences, the task durations can be optimized and task performance (e.g., classification) may potentially be improved. To test our hypothesis, we first retrospectively tracked the changed beta weights generated from general linear models (GLM) from volume to volume on 97 subjects in a stop-signal task. By analyzing the decaying rates of beta weights for various trials and subjects, we were able to determine minimum scan times (MSTs) for each circumstance. The results showed that not only each individual subject produced different MSTs, various trials (specifically, go trials following successful stop trials, go trials following failed stop trials, successful stop trials and failed stop trials) also generated different MSTs, indicating a need of individualization on task durations. We further implemented support vector machine (SVM) for classification on 67 SUD/control labeled subjects and compared the classification accuracies with and without using MSTs. Among 16 classification accuracies with various MSTs, 2 significantly outperformed the accuracies using full trails, indicating an optimizing fMRI performance was achieved by using MSTs. In conclusion, we demonstrated the potentials of an neuroadaptive task design and we believed such methodology can be widely adapted for other task-based, GLM related experiments.
Topic Area: METHODS: Neuroimaging