Automatic choroid plexus assessment in SLE: a deep learning-enabled study

Area of Science:

• Neuroimaging and Artificial Intelligence
• Systemic Lupus Erythematosus (SLE) Pathophysiology
• Cognitive Neuroscience

Background:

• Systemic Lupus Erythematosus (SLE) can affect cognitive function, specifically processing speed and attention deficits (PSAD), even without major neuropsychiatric syndromes.
• The choroid plexus (ChP) role in SLE-related cognitive dysfunction is not fully understood.
• Accurate quantification of ChP volume is crucial for investigating its association with systemic inflammation and cognitive impairments.

Purpose of the Study:

• To develop and validate a deep learning model (Swin-UNETR) for automated choroid plexus segmentation.
• To investigate the relationship between ChP volume, systemic inflammation, and PSAD in SLE patients.
• To identify biomarkers for predicting PSAD risk in SLE.

Main Methods:

• A multicenter retrospective study included 137 SLE patients and 159 healthy controls (HCs).
• The Swin-UNETR model was trained for ChP segmentation on 3D T1-weighted MR images.
• Correlation, mediation, and LASSO regression analyses were performed to assess relationships between ChP volume, systemic inflammation index (SII), complement component 3 (C3), and PSAD.

Main Results:

• Swin-UNETR demonstrated high accuracy in ChP segmentation (median DSC=0.89 internal, 0.82 external).
• SLE patients with PSAD exhibited significantly greater ChP volume compared to SLE patients without PSAD and HCs.
• Increased ChP volume, elevated SII, and low C3 were identified as independent risk factors for PSAD, with ChP volume mediating the SII-PSAD relationship.

Conclusions:

• Enlarged ChP volume serves as a significant biomarker and mediator for PSAD in SLE patients.
• The Swin-UNETR deep learning model provides accurate ChP quantification.
• A three-biomarker panel (ChP volume, C3, SII) offers a practical approach for SLE-PSAD risk stratification.