A computational method to differentiate rheumatoid arthritis patients using thermography data

Area of Science:

• Medical imaging
• Artificial Intelligence
• Rheumatology

Background:

• Traditional rheumatoid arthritis (RA) diagnosis is complex, relying on diverse clinical and imaging data.
• A simplified diagnostic approach integrating blood analysis and thermography is needed.

Purpose of the Study:

• Identify significant clinical predictors for RA diagnosis.
• Determine key thermographic parameters for differentiating RA patients from healthy individuals.
• Evaluate the efficacy of artificial neural networks (ANNs) in distinguishing RA patients.

Main Methods:

• Utilized a dataset of clinical and thermal data from 65 RA patients and 104 healthy subjects.
• Employed univariate logistic regression to identify significant predictors.
• Applied a feedforward neural network model with BFGS optimization and cross-validation for performance assessment.

Main Results:

• The ANN model incorporated 10 parameters, including finger temperature, age, BMI, ESR, CRP, and WBC.
• Achieved high diagnostic performance with 81.25% sensitivity and 100% specificity.
• Overall accuracy of the model reached 92.86%.

Conclusions:

• Thermography shows promise as an adjunct tool for RA screening and diagnosis.
• This methodology can aid in monitoring disease progression alongside existing diagnostic methods.