Prof. Boyko Gueorguiev-Rüegg: “AI-based detection of trauma fractures in the thoracolumbar spine”

Abstract

Fractures of the thoracolumbar spine are frequently occurring but often overseen on plain radiographs leading to pain, deformity and nerve damage. This study investigated the potential of artificial intelligence (AI) based methods to detect such fractures automatically. A total of 630 vertebrae of 151 patients were analyzed based on medical image data. Fracture (N=302) and non-fracture (N=328) outcomes were determined by 3 expert spine surgeons based on computed tomography (CT) and magnetic resonance imaging (MRI) data. Deep learning convolutional neural network models were trained, validated and tested to detect fractures based on sagittal radiographs and the expert-based ground truth. Prediction accuracy was very high (sensitivity of 91% and specificity of 89%), with a heatmap analysis correctly indicating fracture location in 81%. This AI-based tool could be utilized as low-cost effective support for fracture detection on simple radiographs, where more advanced 3D imaging modalities are not available, such as in low-income regions