Finger-aware Artificial Neural Network for predicting arthritis in Patients with hand pain

Abstract

Arthritis is an inflammatory condition associated with joint damage, the incidence of which is increasing worldwide. In severe cases, arthritis can result in the restriction of joint movement, thereby affecting daily activities; as such, early and accurate diagnosis crucial to ensure effective treatment and management. Advances in imaging technologies used for arthritis diagnosis, particularly Single Photon Emission Computed Tomography/Computed Tomography (SPECT/CT), have enabled the quantitative measurement of joint inflammation using ${\text{SUV}}_{\text{max}}$. To the best of our knowledge, this is the first study to apply deep learning to ${\text{SUV}}_{\text{max}}$ to predict the development of hand arthritis. We developed a transformer-based Finger-aware Artificial Neural Network (FANN) to predict arthritis in patients experiencing hand pain, including finger embedding, and to share unique finger-specific information between hands. Compared to conventional machine learning models, the FANN model demonstrated superior performance, achieving an area under the receiver operating characteristic curve of 0.85, accuracy of 0.79, precision of 0.87, recall of 0.79, and F1-score of 0.83. Furthermore, analysis using the SHapley Additive exPlanations (SHAP) algorithm revealed that the FANN predictions were most significantly influenced by the proximal interphalangeal joints of the right hand, in which arthritis is the most clinically prevalent. These findings indicate that the FANN significantly enhances arthritis prediction, representing a promising tool for clinical decision-making in arthritis diagnosis.