Proof of Principle: Inferring three dimensional kinematics of carpal bones in the wrist from fluoroscopic images

The carpal bones ? eight bones that form the wrist ? move in a complex 3D pattern as the hand takes up different positions. Chronic wrist pain arises from several conditions, such as instability patterns, non union or malunion of fractures, osteoarthritis etc. The result for patients is a severe reduction in quality of life due to impairment of everyday functions, lost work time and loss of the capacity to live independently. The current method of distinguishing between these conditions uses fluoroscopy ? a low dose X-ray technique. From the movement of the carpal bones visible on 2D video fluoroscopy sequences taken as the hand moves, experienced clinicians can infer the 3D movements of the carpal bones. Different patterns of movement characterise different conditions. The interpretation is difficult and the diagnostic accuracy depends wholly on the practitioner?s experience. Accurate diagnosis requires referral to a specialist hand consultant, and treatment is often delayed to the detriment of the patient. For research purposes, it is possible to visualise the 3D movement of the carpal bones directly using imaging techniques such as computed tomography or magnetic resonance imaging, but this kind of imaging is not appropriate for clinical investigations.
In this project we will develop computer image analysis programmes that can infer the 3D movements in an individual wrist from the 2D fluoroscopy sequences, with the aim of producing a tool that will allow accurate diagnosis to be performed in the hand clinic without referral for specialist examination. This will improve patient care by achieving earlier diagnosis and treatment and reduce health-care costs. We will use image analysis methods that can ?learn? the range of movements of the carpal bones using CT images collected from normal volunteers, and also learn to identify these movements from corresponding 2D fluoroscopy sequences. Inferring 3D movement from 2D sequences is complicated because, as well as moving in the 2D images, the projected 2D shapes of the bones change as they rotate and the images of the individual bones overlap. Some of the carpal bones have complex shapes, which vary among individuals. This intrinsic shape variation must be taken into account in inferring the movement. We intend to address this problem by using ?statistical shape models? of the carpal bones. These are mathematical descriptions that are also ?learned? from the CT images and which represent the average shape of each bone together with its normal range of variability.

The wrist is a complex joint. While some research has used 3D imaging methods to study the average 3D kinematics of the carpal bones, this has not translated into informing clinical evaluation of wrist pain. Chronic pain in the wrist arises due to a number of conditions, such as instability patterns, non union or malunion of fractures, primary osteoarthritis and inflammatory arthritis. The current method of distinguishing between these conditions is by examining 2D video fluoroscopy sequences showing movement of the hand across its full range in two views. From the patterns of movement of the carpal bones visible on these 2D views experienced clinicians can infer deviations from normal kinematics. The interpretation is difficult and the accuracy of diagnosis depends wholly on the experience of the practitioner. Accurate diagnosis requires referral to a specialist hand consultant and treatment is often delayed to the detriment of the patient. 3D imaging to visualise the movement directly is not appropriate for clinical investigations.
In this project we will develop computer programmes that can infer the 3D kinematics of an individual wrist from the 2D fluoroscopy sequences. Our objective is to create a diagnostic tool that will reduce the subjectivity of the investigation and allow accurate diagnosis in the hand clinic without referral for specialist examination. To achieve this we will develop image analysis software that will ?learn? the range of movements of the carpal bones in normal wrists and be able to identify these from the 2D fluoroscopy sequences. Inferring 3D movement from 2D sequences is complicated because in projection, in addition to moving, the 2D projected shapes of the bones changes as they rotate, and the images of the individual bones overlap. There is intrinsic variation between individuals in the shapes of the bones, which must be taken into account in inferring the movement. We intend to address this problem by using ?statistical shape models? derived from a training set of 3D (CT) images and fluoroscopy sequences showing normal wrist movement. The technical challenge is to extend methods that have been developed for 2D-3D image registration to use 3D statistical models rather than 3D images, and to optimise the registration taking into account natural shape variability. Using the 3D images as ?ground truth? we will evaluate the performance of our programmes for accuracy and robustness in deriving the true 3D movement patterns from the 2D sequences.