Deep Learning Based Automatic Ankle Tenosynovitis Quantification from MRI in Patients with Psoriatic Arthritis: A...

Area of Science:

• Radiology
• Artificial Intelligence
• Rheumatology

Background:

• Tenosynovitis is a common psoriatic arthritis (PsA) manifestation.
• Current assessment relies on semi-quantitative MRI scoring with inherent variability.
• Need for objective, automated quantification of tenosynovitis in PsA.

Purpose of the Study:

• To evaluate a fully automated deep-learning approach for ankle tenosynovitis segmentation and volume quantification from MRI in PsA patients.
• To compare automated volumetric measurements with traditional radiologist scoring.
• To assess the utility of automated quantification in detecting disease progression over time.

Main Methods:

• Analysis of 364 ankle 3T MRI scans from 71 PsA patients.
• Development of a deep-learning segmentation model (nnUNet framework) using manually annotated ground truth.
• Validation using Dice scores for segmentation accuracy and Spearman correlation for comparing volumetric and visual scores.

Main Results:

• The deep-learning model achieved high performance with mean Dice scores between 0.84 and 0.92.
• Pathology volumes showed significant correlation with visual scores (Spearman ρ = 0.52-0.62).
• Automated quantification detected subtle inflammatory changes and disease progression not evident in semi-quantitative scores.

Conclusions:

• An automated segmentation tool provides fast and accurate ankle tenosynovitis quantification in PsA.
• This approach enhances sensitivity to disease progression.
• Volume-based metrics can complement visual scoring for comprehensive PsA assessment.