AI-based X-ray fracture analysis of the distal radius: accuracy between representative classification, detection and segmentation deep learning models for clinical practice

Abstract

ObjectivesTo aid in selecting the optimal artificial intelligence (AI) solution for clinical application, we directly compared performances of selected representative custom-trained or commercial classification, detection and segmentation models for fracture detection on musculoskeletal radiographs of the distal radius by aligning their outputs.Design and settingThis single-centre retrospective study was conducted on a random subset of emergency department radiographs from 2008 to 2018 of the distal radius in Germany.Materials and methodsAn image set was created to be compatible with training and testing classification and segmentation models by annotating examinations for fractures and overlaying fracture masks, if applicable. Representative classification and segmentation models were trained on 80% of the data. After output binarisation, their derived fracture detection performances as well as that of a standard commercially available solution were compared on the remaining X-rays (20%) using mainly accuracy and area under the receiver operating characteristic (AUROC).ResultsA total of 2856 examinations with 712 (24.9%) fractures were included in the analysis. Accuracies reached up to 0.97 for the classification model, 0.94 for the segmentation model and 0.95 for BoneView. Cohen’s kappa was at least 0.80 in pairwise comparisons, while Fleiss’ kappa was 0.83 for all models. Fracture predictions were visualised with all three methods at different levels of detail, ranking from downsampled image region for classification over bounding box for detection to single pixel-level delineation for segmentation.ConclusionsAll three investigated approaches reached high performances for detection of distal radius fractures with simple preprocessing and postprocessing protocols on the custom-trained models. Despite their underlying structural differences, selection of one’s fracture analysis AI tool in the frame of this study reduces to the desired flavour of automation: automated classification, AI-assisted manual fracture reading or minimised false negatives.

Artificial Intelligence in Emergency Trauma Care: A Preliminary Scoping Review

“…Convolutional neural networks attained over 90% sensitivity and specificity for detecting solid organ abdominal trauma like spleen, liver, and kidney lesions on CT scans. 9 Additional models achieved up to 97% accuracy in diagnosing distal radius fractures on radiographs, 11 98% sensitivity for hip fractures on pelvic X-rays, 14 and AUC exceeding 0.80 for intracranial hemorrhage detection on head CT scans. 19 Deep learning also shows precision in localizing traumatic findings, with activation mapping techniques precisely pinpointing 95.9% of hip fracture lesions 14 and models consistently highlighting displaced ribs on chest CTs.…”

“… The review underscores the necessity for a hybrid, user-tailored CDSS that can interface with oral, video, and digital data, emphasizing iterative evaluation of CDSS’s intrinsic characteristics and their clinical impact throughout the IT lifecycle. Triage [ 11 ] Russe, Rebmann, Tran, Kellner, Reisert, Bamberg, Kotter, Kim AI-based X-ray fracture analysis of the distal radius: accuracy between representative classification, detection and segmentation deep learning models for clinical practice. Comparative Observational Germany Evaluated AI models for detecting distal radius fractures in radiographs, comparing custom-trained classification, detection, segmentation models, and a commercial solution, showing high accuracies (up to 0.97) in fracture detection.…”

“…Most of the investigated studies demonstrated that CNNs can achieve exceptional diagnostic accuracy, with sensitivity and specificity exceeding 90% for diverse traumatic injuries identifiable on CT, X-ray, and MRI. 9 , 11 , 14 , 19 , 24 However, reliance on single-center, retrospective data risks optimism bias and overfitting, requiring further multi-institutional prospective validation encompassing diverse patient populations and scanners before responsible clinical integration. 9 , 11 , 14 , 19 , 21 , 24 Additional gaps emerge in standardized injury quantification and localization.…”

“… 9 , 11 , 14 , 19 , 24 However, reliance on single-center, retrospective data risks optimism bias and overfitting, requiring further multi-institutional prospective validation encompassing diverse patient populations and scanners before responsible clinical integration. 9 , 11 , 14 , 19 , 21 , 24 Additional gaps emerge in standardized injury quantification and localization. Injury severity scoring models demonstrate comparable performance to validated classifications like ISS but currently suffer from mortality overestimation, requiring refinements to improve calibration.…”