Gout Diagnosis From Ultrasound Images Using a Patch-Wise Attention Deep Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Rheumatology

Background:

• Gout prevalence is increasing globally, requiring improved diagnostic methods.
• Ultrasonography is a valuable tool for gout diagnosis due to its non-invasive nature and cost-effectiveness.
• Current diagnostic approaches can be enhanced by automated analysis of ultrasound images.

Purpose of the Study:

• To develop and validate a deep learning-based artificial intelligence (AI) model for automated gout diagnosis.
• To utilize ultrasound images of the first metatarsophalangeal joint (MTP1) for AI model training and validation.
• To enhance the detection of subtle sonographic features indicative of gout.

Main Methods:

• A deep learning model with patch-wise attention and multi-scale feature extraction was developed.
• Ultrasound images from 598 cases across two institutions were used.
• The model was trained on data from Institution 1 and validated internally and externally using data from Institution 2.

Main Results:

• The AI model achieved high diagnostic performance: 87.88% accuracy, 87.85% sensitivity, 87.93% specificity, and 93.43% AUC.
• The model generates interpretable heatmaps to pinpoint gout-related pathological features.
• These heatmaps aid in clinical decision-making by localizing abnormalities.

Conclusions:

• A novel AI model for automated gout detection using ultrasound images was successfully developed.
• The model demonstrated superior performance compared to existing methods.
• The AI model's highlighted features correlate with expert assessments, indicating its potential as a diagnostic aid.