Using machine learning to identify screw-rod systems and support clinical decision-making

For the first time, our research group attended an event of the Deutsche Gesellschaft für Neurochirurgie. Richard Klein presented a AI-based solution for identifying screw-rod systems from radiological imaging data at this year’s section days in Münster. The project was initiated by the Department of Neurosurgery at the Universitätsklinikum Jena, which supported the initiative with expert guidance and an extensive dataset. The presentation not only provided insights into current research findings but also demonstrated the practical implementation of an interdisciplinary approach bridging medical technology and clinical application.

This system addresses a central challenge in neurosurgical practice. When specifications of implanted systems are incomplete or unavailable, clinicians face significant difficulties in preoperative planning. This issue becomes particularly critical in revision surgeries or emergency situations, where rapid and reliable information about existing implants is essential. Accurate identification therefore remains a key prerequisite for safe and efficient patient management. The presented prototype applies modern machine learning techniques to close this information gap. It extracts characteristic features from radiological images and uses them for classification. Initial results show class-specific sensitivity of up to 97.1 percent when distinguishing between four different manufacturers. These findings highlight the potential of AI-based approaches to support clinical decision-making and streamline surgical workflows.

Beyond the presentation, the event enabled intensive scientific exchange with experts from both clinical practice and research and provided valuable insights into current challenges in neurosurgery. The feedback and discussions will directly inform the further development of the project. The next steps focus on expanding the underlying dataset to improve the robustness and predictive power of the models. In addition, the team plans to integrate further imaging modalities to enhance generalizability and enable prospective use in routine clinical practice.