New Image Processing Algorithm for ROI Extraction in Patients with Rheumatoid Arthritis

This study suggested a novel algorithm for extracting region of interest (ROI) to measure the joint space widths (JSWs) from patients with rheumatoid arthritis (RA) using radiographic image of hands. The radiographic image was first preprocessed by three steps, including noise reduction, diffusion and segmentation. Then, phalangeal regions (PRs) corresponding to the bone structures of each finger were extracted by applying step-wedge functions to the preprocessed image. After extraction of PRs, proper branch path of phalanx (BPP) was also extracted. Each of the five extracted BPPs exactly matched the bone structures of each finger and runs through the center of each finger. Therefore, the algorithm automatically detected 14 joints identified as sharp changes in gray scale intensity along BPP through the profile plot. And the ROIs corresponding to 14 joints were extracted. JSWs of each joint were manually measured within these ROIs. The 30 radiographic images from three groups (10 images each) were tested. The performance was evaluated by measuring joint location percentage errors and mean JSWs for distal interphalangeal (DIP), proximal interphalangeal (PIP), interphalangeal (IP), and metacarpophalangeal (MCP) joints. The novel algorithm succeeded in detecting correctly 96.90% of total joints and failed to detect 6 joints among 14 joints in RA patient with severe deformity. Furthermore, the mean JSWs of normal joints in both hands were 0.35-1.66 mm greater than those of RA patients.