Utilizing Optimized Mixed-Order Relation-Aware Recurrent Neural Network for Metacarpophalangeal Rheumatoid Arthritis...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Rheumatology
• Ultrasound Diagnostics

Background:

• Metacarpophalangeal rheumatoid arthritis (RA) diagnosis relies heavily on sonographer skill.
• Current grading systems evaluate bone and synovium features, but subjectivity remains a challenge.
• Objective and automated grading of RA via ultrasound is needed.

Purpose of the Study:

• To propose an optimized deep learning model for grading metacarpophalangeal rheumatoid arthritis (RA) using ultrasound images.
• To improve the accuracy and consistency of RA grading by automating feature extraction and evaluation.
• To develop a robust method for identifying synovium thickening and bone deterioration.

Main Methods:

• Utilized an optimized mixed-order relation-aware recurrent neural network (MORARNN) for RA grading (MRAG-UI-MORARNN-BWKA).
• Employed confidence partitioning sampling filtering (CPSF) for image pre-processing and unpaired multi-view graph clustering (UMGC) for region segmentation.
• Integrated holistic dynamic frequency transformer (HDFT) for geometric feature extraction and the Black winged kite algorithm (BWKA) for model optimization.

Main Results:

• The proposed MRAG-UI-MORARNN-BWKA method achieved high performance metrics.
• Achieved an accuracy of 97.02%, precision of 97.5%, and sensitivity of 97.25% in grading RA.
• Demonstrated significant improvements and robustness compared to existing diagnostic methods.

Conclusions:

• The developed AI model offers a significant advancement in the automated grading of rheumatoid arthritis from ultrasound images.
• The MRAG-UI-MORARNN-BWKA method provides a more objective and reliable diagnostic tool for clinicians.
• This approach enhances the evaluation of synovium thickening and bone erosion in metacarpophalangeal joints.