Hip joint space analysis for osteoarthritis diagnosis

Abstract

The hip joint is one of the body’s largest weight-bearing joints and is commonly affected by osteoarthritis (OA). With hip osteoarthritis the cartilage within the joint begins to break down and the underlying bone begins to change. The major pathological features for hip OA include joint space narrowing and osteophytes formation. The current gold standard for diagnosing OA is plain radiography, however, is insensitive when detecting early OA changes and it depends on the subjectivity of the practitioner. The development of machine learning algorithms which help in the study and diagnosis of many medical concerns has made huge progress. In this paper a computer-aided method to facilitate OA diagnosis from DXA images is proposed. Two different datasets were provided and two different methods were developed for each of them. For the first data set, Unet Convolutional Neural Network was applied to automatically segment the hip joint space. Moreover, a quantitative analysis of the joint space width (JSW) was performed to study the relationship between joint space width and Anthropometric parameters. On the second dataset, with known conditions of osteoarthritis, images were classified with VGG16 neural network. Results of the automatic segmentation had a Dice Coefficient of 0.835. Inverse relationships between joint space narrowing and age were found and differences between men and women were significant. Regarding the second dataset, images were classified with a precision of 0.975 and a recall of 0.985. While the approach is a step in the direction of OA diagnosis, developing a tool that is accurate enough to be applicable in a clinical context still remains an open challenge.