[Artificial intelligence in fracture diagnostics : Potentials and challenges in the clinical practice].
Authors
Affiliations (2)
Affiliations (2)
- Klinik für Diagnostische und Interventionelle Radiologie, Universitätsklinikum Bonn, Venusberg-Campus 1, 53127, Bonn, Deutschland. [email protected].
- Klinik für Diagnostische und Interventionelle Radiologie, Universitätsklinikum Bonn, Venusberg-Campus 1, 53127, Bonn, Deutschland.
Abstract
Trauma surgery and emergency medical care are facing growing challenges: rising patient numbers, a shortage of specialists and a high diagnostic workload are leading to diagnostic errors, up to 80% of which relate to overlooked fractures. Systems based on artificial intelligence (AI) for fracture diagnostics offer promising support in this context. Modern deep learning algorithms, in particular convolutional neural networks, achieve high sensitivities and specificities in the detection of frequent fractures in large validation studies. As a "second reader", they increase diagnostic accuracy, reduce diagnostic time and improve patient safety, especially in the case of subtle fractures or limited practitioner experience. Additional applications include automated triage, angle measurements, bone age determination and the detection of other pathologies. Limitations include heterogeneous training data, limited performance in complex fractures and regulatory requirements. The continuous technological development promises increasing performance and broader fields of application for AI in fracture diagnostics. Future systems will also increasingly enable multimodal and 3D analyses as well as deeper integration into the clinical workflow. The use of AI does not replace physicians but acts as an assistive tool to increase quality and efficiency; however, further independent, prospective and patient-orientated studies and integration into clinical guidelines are required for a broad implementation.