SdADC: Semi-supervised Diffusion MRI Artefact Detection and Classification

Motivation: Diffusion magnetic resonance imaging (dMRI) is prone to artefacts, which can significantly impact the preprocessing and downstream analysis. Goal(s): Develop automatic methods to detect and classify the dMRI artefacts to exclude problematic cases for further analysis. Approach: A two-stage deep learning-based framework is proposed to detect the artefacts using angular resolution enhanced fractional anisotropy (FA) and then classify the specific type of artefacts. Results: The proposed method shows consistently good performance in dMRI artefact detection and classification across HCP and PPMI datasets. Impact: Our method improves dMRI data reliability by automating artifact detection and classification using a two-stage deep learning approach with angular resolution-enhanced FA. The propsoed framework consistently identifies and categorizes artifacts, enhancing preprocessing and analysis across large-scale diffusion MRI datasets.