A Sub-Pixel Accurate Quantification of Joint Space Narrowing Progression in Rheumatoid Arthritis

• Published in: IEEE journal of biomedical and health informatics Vol. 27; no. 1; pp. 53 - 64
• Main Authors: Ou, Yafei, Ambalathankandy, Prasoon, Furuya, Ryunosuke, Kawada, Seiya, Zeng, Tianyu, An, Yujie, Kamishima, Tamotsu, Tamura, Kenichi, Ikebe, Masayuki
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.01.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Arthritis; Arthritis, Rheumatoid - diagnostic imaging; Autoimmune diseases; Bones; Computer-aided Diagnosis; Diagnostic radiography; Disease Progression; Error analysis; Error reduction; Finger Joint; Frequency Domain Analysis; Humans; Image registration; Image segmentation; Joint diseases; Joint Space Narrowing; Joints (anatomy); Medical diagnosis; Medical imaging; Monitoring; Phantom Imaging; Pixels; Radiography; Radiology; Rheumatoid Arthritis; Sensitivity; Spatial discrimination; Spatial resolution; Standard deviation; Synovial joints; Wrist
• ISSN: 2168-2194; 2168-2208
• DOI: 10.1109/JBHI.2022.3217685

Rheumatoid arthritis (RA) is a chronic autoimmune disease that primarily affects peripheral synovial joints, like fingers, wrists and feet. Radiology plays a critical role in the diagnosis and monitoring of RA. Limited by the current spatial resolution of radiographic imaging, joint space narrowing (JSN) progression of RA for the same reason above can be less than one pixel per year with universal spatial resolution. Insensitive monitoring of JSN can hinder the radiologist/rheumatologist from making a proper and timely clinical judgment. In this paper, we propose a novel and sensitive method that we call partial image phase-only correlation which aims to automatically quantify JSN progression in the early RA. The majority of the current literature utilizes the mean error, root-mean-square deviation and standard deviation to report the accuracy at pixel level. Our work measures JSN progression between a baseline and its follow-up finger joint images by using the phase spectrum in the frequency domain. Using this study, the mean error can be reduced to 0.0130 mm when applied to phantom radiographs with ground truth, and 0.0519 mm standard deviation for clinical radiography. With the sub-pixel accuracy far beyond usual manual measurements, we are optimistic that the proposed work is a promising scheme for automatically quantifying JSN progression.