Interpretable Deep Learning(DL) models are the next step in establishing DL prediction models as accepted tools that provide researchers with data-driven methods to further understand neuroimaging data. In this work, we developed two interpretable DL models to predict Working Memory(WM) scores from task fMRI data to assess neural circuitry pertaining to WM; wherein a traditional Convolutional Neural Network(CNN)(1-3) contained fMRI activation data from cortical vertices as a single image, and the second contained cortical activation data from both hemispheres as separate channels. Overall, the interpretable DL model provided high quality saliency maps potentially displaying novel regions pertaining to WM.
This abstract and the presentation materials are available to members only; a login is required.