Artificial intelligence for automated detection and measurements of carpal instability signs on conventional radiographs

Hendrix, Nils

DOI: https://doi.org/10.1007/s00330-024-10744-1

IF: 7.034

2024-04-18

European Radiology

Abstract:To develop and validate an artificial intelligence (AI) system for measuring and detecting signs of carpal instability on conventional radiographs.

radiology, nuclear medicine & medical imaging

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve

This paper aims to develop and validate an artificial intelligence (AI) system for the automatic detection and measurement of signs of wrist instability on routine X-rays. Specifically, the study has two objectives:

1. **Develop and Validate the AI System**: Develop an AI system capable of accurately measuring and detecting signs of wrist instability, including the measurement of scapholunate joint distance, scapholunate angle, and capitolunate angle, as well as the detection of wrist arc disruptions.
2. **Comparison with Clinicians**: Evaluate the performance of the AI system in measuring and detecting signs of wrist instability on routine X-rays, compared to five clinicians with different professional backgrounds.

### Background

Wrist instability refers to the inability of the carpal bones to maintain normal alignment under physiological load and motion. Most wrist instabilities are caused by acute trauma, such as ligament tears and bone displacements. Non-traumatic causes are less common and include inflammatory arthritis, infections, and congenital diseases. Although the true prevalence of wrist instability is unclear, studies suggest that acute wrist fractures are often accompanied by traumatic ligament injuries. For example, surgical validation data show that the incidence of scapholunate ligament tears is 34% in all scaphoid fractures and 13%-64% in all distal radius fractures. Early identification of these injuries is crucial because untreated cases may lead to scapholunate dissociation and scapholunate advanced collapse (SLAC). However, signs of wrist instability on routine X-rays are often overlooked. Therefore, the focus of the study is on detecting signs of wrist instability through routine X-rays.

### Methods

The study used two case-control datasets from three hospitals in the Netherlands (Hospitals A, B, and C). Dataset 1 (2,178 X-rays from 1,993 patients collected between 2018-2019) was used to develop the AI system, and Dataset 2 (481 X-rays from 217 patients collected between 2017-2021) was used for testing. The study also included an observer study to compare the performance of the AI system with that of five clinicians.

### Results

- **Measurement Accuracy**: The AI system's mean absolute errors in measuring scapholunate joint distance, scapholunate angle, and capitolunate angle were 0.65 mm, 7.9 degrees, and 5.9 degrees, respectively.
- **Detection Accuracy**: The AI system's sensitivity and specificity in detecting wrist arc disruptions were 83% and 64%, respectively.
- **Comparison with Clinicians**: The AI system's measurement results were roughly equivalent to those of the clinicians, but it was more accurate than most clinicians in detecting wrist arc disruptions.

### Conclusion

The study demonstrates that the newly developed automated AI system can accurately measure and detect signs of wrist instability on routine X-rays. This system has the potential to improve the detection of wrist arc disruptions and may become a useful tool to support clinicians in detecting wrist instability.