Deep learning QC

Challenge

Solution

Train a deep learning model to predict QC scores from the imaging alone

Convolutional neural network architecture (part 1)
Convolutional neural network architecture (part 2)

Deep learning loss curves

Loss stabilizes after ~20 epochs, for both and imaging only model and a model with imaging and QC metric input.

  • without QSIPrep metrics
Deep learning loss curve without auto-QC metrics
  • with QSIPrep metrics
Deep learning loss curve with auto-QC metrics
Deep learning model ROC curve.