AI system accurately detects and measures signs of carpal instability on conventional wrist radiographs

Researchers from Radboudumc have developed a novel artificial intelligence (AI) system that precisely detects and measures signs of carpal instability in conventional wrist radiographs. The study, led by Nils Hendrix of the Department of Medical Imaging, was published in European Radiology.

Carpal instability occurs when the wrist bones lose proper alignment, often due to trauma like ligament tears or fractures. Conventional radiography is the standard first-line imaging technique for suspected wrist injuries, but subtle indicators of carpal instability are often missed. The AI system aims to enhance detection and measurement of these subtle signs, supporting clinicians in making more accurate diagnoses. Previously, the team developed an AI system for automated detection of scaphoid fractures (see: link1, link2), a condition that frequently co-occurs with carpal instability. These AI systems could complement each other, enhancing the overall diagnostic process for wrist trauma.

The research team consisted of experts from Radboudumc, along with collaborators from Jeroen Bosch Hospital, Hospital Gelderse Vallei, and the Jheronimus Academy of Data Science, including Ward Hendrix, Bas Maresch, Job van Amersfoort, Tineke Oosterveld-Bonsma, Stephanie Kolderman, Myrthe Vestering, Stephanie Zielinski, Karlijn Rutten, Jan Dammeier, Lee-Ling Sharon Ong, Bram van Ginneken, and Matthieu Rutten.

The researchers trained and tested the AI system using two datasets of wrist radiographs collected from three hospitals. The AI was programmed to assess key indicators of carpal instability, such as scapholunate (SL) distances, scapholunate and capitolunate (CL) angles, and interruptions in carpal arcs. Its performance was benchmarked against five clinicians from different specialties.

The AI system demonstrated comparable accuracy to clinicians in measuring SL distances and SL and CL angles, and it surpassed most clinicians in detecting carpal arc interruptions. Specifically, it achieved 83% sensitivity and 64% specificity for detecting arc interruptions, showcasing its potential as a valuable tool, especially for clinicians less familiar with this complex diagnosis.

This AI system could potentially aid in the diagnosis of carpal instability in clinical practice by improving diagnostic accuracy and efficiency while reducing the burden on radiologists. The investigated methods could be useful for automating other carpal instability measurements and measurements in other musculoskeletal structures. Ongoing research will focus on clinical validation and exploring the system's potential to enhance patient outcomes.

Read the full study here

Hendrix N, Hendrix W, Maresch B, et al. Artificial intelligence for automated detection and measurements of carpal instability signs on conventional radiographs. Eur Radiol.2024;34:6600–6613. https://doi.org/10.1007/s00330-024-10744-1.