Abstract:
HTML
Purpose: Rheumatoid arthritis (RA) of the hand is a prevalent and costly healthcare problem affecting approximately 1% of the population. Radiography is considered the gold standard for RA outcome measures since it provides visualization of structural changes. RA affects the interarticular cartilage causing narrowing of the joint space width (JSW) in the adjacent bones, which is visible radiographically. Currently changes in JSW are assessed using subjective semiquantiative scoring systems that do not directly measure the distance between the joint margins. We present here an artificial neural network (ANN) based software algorithm to quantify JSW in joints of the wrist that augments previous work to measure JSW in the distal hand. The software automatically measures JSW in 6 joints of the wrist: the carpometacarpal (CMC) joints for digits 3 and 5, the radiolunate, radioscaphoid, scaphocapitate, and lunatocapitate joints.
Methods and Materials: The algorithm employed an ANN, that was trained using a jackknife technique, and tested on a data set of 30 wrist radiographs from subjects with mild to moderate RA. This provided a total of 180 joints for evaluation. An initial edge detection step identified candidate points for the proximal and distal margins of each joint. The ANN was used to determine which points lay on the true joint margins. A gold standard JSW (JSWGS) was calculated for each joint using hand delineated joint margins. We evaluated the software performance by determining the R2 values from a linear regression fit, and by calculating RMSDev = sqrt{Σ(JSW-JSWGS)2/(180-NFail)}, where NFail was the total number of cases where the software was unable to find a JSW value. We also calculated the success ratio as (180-NFail)/180.
Results: We measured R2 = 0.79 and RMSDev = 0.25 mm. We found NFail = 8, giving a success ratio of 96%. The CMC joints presented the greatest challenge to the software due to overlapping and ambiguous margins. In practice, the algorithms failures will be handled using a manual correction step.
Conclusion: This work represents a significant step in our goal of developing an automated computerized system to fully assess hand/wrist radiographs for structural changes due to RA. This work should provide a more objective and quantitative method to assess joint damage in RA of the wrist. This research was supported by a grant from the Whitaker Foundation.
Questions about this event email: jduryea@bwh.harvard.edu
Duryea PhD, J,
Neural Network Based Algorithm To Quantify Joint Space Width on Wrist Radiographs for Arthritis Assessment. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3102114.html
