A recent study published in Radiography examined three AI tools for fracture detection in x-rays: BoneView (Gleamer), Rayvolve (AZmed), and RBfracture (Radiobotics). Conducted by lead author Ina Luiken, MD, at TUM University, the analysis involved 1,037 patients and highlighted the strengths and limitations of each model. While all demonstrated moderate to high performance for straightforward fractures, they fell short in complex situations, such as multiple fractures and dislocations. Rayvolve showed the highest sensitivity, BoneView provided balanced performance, and RBfracture excelled in specificity. None exceeded 91% accuracy for acute fractures. The study emphasizes that current AI algorithms should assist rather than replace radiologists. The findings indicate the potential for improving clinical practice, suggesting Rayvolve for initial screenings, BoneView as a second-reader tool, and RBfracture for ruling out fractures. The researchers called for multicenter validation and diverse training data to enhance the generalizability of these AI models.
