When dealing with very large imaging datasets, it becomes hard to perform quality control (QC) of the data collected. Several solutions have been proposed in literature including the use of Artificial Intelligence (AI) to perform automated QC of MRI and EEG on both single-site and multi-site datasets.
Given the successful studies using Convolutional Neural Networks (CNNs) when working with medical images, we trained a CNN – DenseNet – to test whether an automated QC pipeline for brain [18F]-FDOPA PET imaging as a biomarker for the dopamine system could be implemented to return the expected output classification with a minimal error.
Particularly, we trained DenseNet for two different classification problems: the assessment of the image alignment to a standard template and the assessment of the SNR of the images compared to those that were manually-assessed. Both CNNs returned a high accuracy on the training and the independent test datasets. The method is far from clinical use, but these encouraging results support the application of deep learning for automated quality control of medical imaging data.
Ref: Pontoriero A et al, “Automated Data Quality Control in FDOPA brain PET Imaging using Deep Learning”, CMPB (2021)