Identifying scapholunate ligamentous injury

Abstract

The first purpose of this study was to develop a noninvasive clinical tool that could predict whether the scapholunate interosseous ligament and other secondary stabilizing ligaments are injured in the presence of suspected scapholunate instability. The second purpose of this study was to determine which of those ligaments or ligament groups have been injured. Kinematic and three-dimensional (3D) meaurements from 62 cadaver wrists moved in a wrist joint motion simulator were used to develop various neural network predictive models. One group of models was based on angular changes in scaphoid and lunate motion before and after ligament sectioning (representing scapholunate instability). A second group of models was based on changes in the minimum distance between the scaphoid and lunate as well as other 3D gap measurements. The models, based on the scaphoid and lunate angular data, could predict with a 93% accuracy rate whether the wrist ligaments were intact. These models could also predict whether it was the dorsal ligaments or the volar ligaments that were sectioned 84% of the time. The models worked best using data with the wrist in 10 to 30 degrees of wrist flexion. The viability of a CT-based predictive model has been demonstrated by obtaining high prediction rates, sensitivity, specificity, and kappa statistic values.

Figure 1

Two of the gap measurements used in the gap neural network model: dorsal and volar gap openings between the scaphoid and lunate.

Figure 2

Effect of the wrist flexion-extension position on the neural network model’s ability to predict the integrity of wrist ligaments. Here the angle model has the highest correct prediction rate at 10 or 30 degrees of wrist flexion. For this model, the spread value was 0.2.

Figure 3

Effect of the wrist radioulnar deviation position on the neural network model’s ability to predict the integrity of wrist ligaments. Here the angle model has the highest correct prediction rate at 0 degrees of wrist radioulnar deviation. For this model, the spread value was 0.4.

Figure 4

Effect of the neural network spread variable on tuning the model, with the wrist at 30 degrees of flexion. Shown are the results for the angle model when trying to predict the integrity of the wrist ligaments.