3D-QCNet - A pipeline for automated artifact detection in diffusion MRI images

Abstract

Artifacts are a common occurrence in Diffusion MRI (dMRI) scans. Identifying and removing them is essential to ensure the accuracy and viability of any post-processing carried out on these scans. This makes quality control (QC) a crucial first step prior to any analysis of dMRI data. Several QC methods for artifact detection exist, however they suffer from problems like requiring manual intervention and the inability to generalize across different artifacts and datasets. In this paper, we propose an automated deep learning (DL) pipeline that utilizes a 3D-Densenet architecture to train a model on diffusion volumes for automatic artifact detection. Our method is validated on 9000 volumes sourced from 7 large clinical datasets spanning different acquisition protocols (with different gradient directions, high and low b-values, single-shell and multi-shell acquisitions) from multiple scanners. Additionally, they represent diverse subject demographics including age, sex and the presence or absence of pathologies. Our QC method is found to accurately generalize across this heterogenous data by correctly detecting 92% artifacts on average across our test set. This consistent performance over diverse datasets underlines the generalizability of our method, which currently is a significant barrier hindering the widespread adoption of automated QC techniques. Thus, 3D-QCNet can be integrated into diffusion pipelines to effectively automate the arduous and time-intensive process of artifact detection.

Keywords: Artifacts; Deep learning; Diffusion MRI; MRI; Quality control.

Fig. A2.

Precision Recall Curve with thresholds visualized for 3D-QCNet applied to Dataset 7.

Fig. 1.

3D-QCnet DenseNet Model Architecture showing how a 3D volume is processed by a series of dense and transition blocks before being sent to the GAP and dense layers for final classification into the two classes – artifact and normal.

Fig. 2.

ROC curves for Test Datasets.

Fig. 3.

Scans from the test set illustrated to demonstrate 3D-QCNet’s model performance with respect to ground-truth. True Positive samples – A Ghosting artifact, B Herringbone artifact, C Motion/interslice instability artifact, D Faint Chemical artifact (marked in yellow). True Negative samples – E Weighted Image is noisy but is correctly marked as normal. F B0 image with no artifacts. False Positive samples – G Abnormal anatomy of the brain may be affecting the classifier. H Weighted image is noisy but there are no visible artifacts; the model may be too sensitive. False Negative samples – I Chemical shift artifact alongside instability and susceptibility.

Fig. 4.

These scans are from a volume that was marked as normal by our annotator but 3D-QCNet labelled it as having an artifact. Later, on closer inspection it was found to have ghosting artifacts in the ventricles and subcortical regions along with some interslice instability.

Fig. 5.

This volume was marked as having an artifact by 3D-QCNet however our expert human rater (DP) believes that while there is some susceptibility present, it is not enough to warrant removing the data. In such borderline cases, being able to control the model’s sensitivity through probability thresholds, will allow users to fine-tune results based on their preferences.