Objectives: To evaluate how different test set sampling strategies—random selection and balanced sampling—affect the performance of artificial intelligence (AI) models in pediatric wrist fracture detection using radiographs, aiming to highlight the need for standardization in test set design. Materials and methods: This retrospective study utilized the open-sourced GRAZPEDWRI-DX dataset of 6091 pediatric wrist radiographs. Two test sets, each containing 4588 images, were constructed: one using a balanced approach based on case difficulty, projection type, and fracture presence and the other a random selection. EfficientNet and YOLOv11 models were trained and validated on 18,762 radiographs and tested on both sets. Binary classification and object detection tasks were evaluated using metrics such as precision, recall, F1 score, AP50, and AP50-95. Statistical comparisons between test sets were performed using nonparametric tests. Results: Performance metrics significantly decreased in the balanced test set with more challenging cases. For example, the precision for YOLOv11 models decreased from 0.95 in the random set to 0.83 in the balanced set. Similar trends were observed for recall, accuracy, and F1 score, indicating that models trained on easy-to-recognize cases performed poorly on more complex ones. These results were consistent across all model variants tested. Conclusion: AI models for pediatric wrist fracture detection exhibit reduced performance when tested on balanced datasets containing more difficult cases, compared to randomly selected cases. This highlights the importance of constructing representative and standardized test sets that account for clinical complexity to ensure robust AI performance in real-world settings. Key Points: Question Do different sampling strategies based on samples’ complexity have an influence in deep learning models’ performance in fracture detection? Findings AI performance in pediatric wrist fracture detection significantly drops when tested on balanced datasets with more challenging cases, compared to randomly selected cases. Clinical relevance Without standardized and validated test datasets for AI that reflect clinical complexities, performance metrics may be overestimated, limiting the utility of AI in real-world settings. Graphical

Artificial intelligence performance in pediatric fracture detection was influenced by test set composition and image complexity, where an internal test set of difficult radiographs was associated with decreased odds of correct predictions.

Background: Artificial Intelligence (AI) in radiology has shown promise in detecting fractures on initial X-rays. However, the role of follow-up examinations in enhancing AI performance remains unexplored. This study evaluates the impact of including follow-up X-rays on the performance of neural networks in detecting pediatric wrist fractures. Methods: Using the publicly available GRAZPEDWRI-DX dataset of 20,327 pediatric wrist X-rays, we created four training datasets: initial X-rays alone and combinations with follow-up X-rays (with and without casts). Two neural networks, EfficientNet (image classification) and YOLOv8 (object detection), were trained and evaluated using precision, recall, F1 score, and AP metrics. The dataset was divided into training, validation, and test sets, with 500 initial X-rays separated and reserved for testing. Results: EfficientNet models showed no statistically significant improvements in classification performance with the inclusion of follow-up X-rays. In contrast, YOLOv8 demonstrated improved object detection metrics, particularly AP50 ( p = 0.003) and F1 score ( p = 0.009), when follow-up X-rays were included. The improvement was most evident when both cast and non-cast follow-ups were incorporated. Conclusion: Adding follow-up X-rays did not enhance classification performance but improved fracture localization in object detection tasks. These findings suggest that including follow-up data shows no relevant improvement in the detection rate of fractures but can enhance AI applications for pediatric wrist fracture detection, particularly for object detection models.

The aim of our systematic work was to investigate an artificial intelligence-based prediction of relevant internal injuries of the paediatric knee joint ba

Despite growing clinical adoption of artificial intelligence in radiology, fewer than one in five commercially available models are cleared for paediatric use. Existing systems are trained predominantly on adult data and generalize poorly to neonates, yet efficient strategies for repurposing them without data sharing or large-scale local data curation are lacking. Here we present an international retrospective study of neonatal chest radiographs from 30 neonatal intensive care units across 20 countries and territories, evaluating continual learning as a privacy-preserving approach to adapt a deep learning model for adult endotracheal tube assessment to neonatal care. Each hospital contributed 50 radiographs for sequential model retraining without centralized data pooling. Continual learning achieved lower anatomical landmark localization error (0.67 cm versus 1.29 cm for hospital-specific fine-tuning and 4.30 cm for adult-only models) at 29 of 30 sites, with improved detection of deep intubations at 27 of 30 sites. These findings demonstrate that privacy-preserving collaborative learning can address critical gaps in paediatric AI deployment.

Summary The reporting rate of vertebral fractures (VFs) identified on routine thoracic and abdominal CT scans was evaluated. Only 14.7% of VFs were documented in the impression section of the radiology report. These findings indicate that the recognition and reporting of VFs on routine CT examinations have not improved over the past decade. Purpose Vertebral fractures (VFs) are the most common osteoporotic fractures, yet they are substantially underreported on routine computed tomography (CT) scans. This study aimed to quantify the detection and reporting rates of VFs on thoracic and abdominal CT scans performed for non-musculoskeletal indications. Methods Of 8500 CT scans performed at a large Austrian tertiary-care hospital in 2023, 1500 thoracic and abdominal CT scans were randomly selected. Two musculoskeletal radiologists independently assessed vertebral bodies of the thoracic and lumbar spine using Genant’s semiquantitative method. Vertebral deformities were categorised as Grade 1 (20–25% height loss), Grade 2 (25–40%), or Grade 3 (>40%). For the purpose of this study, a VF was defined as Grade ≥ 2. Findings were compared with the original radiology reports to determine detection and reporting rates. Results Consensus results were available for 1481 CT scans from 1364 individuals. Overall, 170 VFs (11.5%) were identified in 164 patients. Of these, 35.9% (n = 61) were lumbar, 52.4% (n = 89) were thoracic, and 11.8% (n = 20) involved both regions. Only 14.7% of all VFs were documented in the impression section of the corresponding radiology report, whereas 50.0% (85/170) were mentioned in the narrative interpretation section. Inter-reader agreement was excellent (κ = 0.94). Conclusion Despite increasing awareness of osteoporosis, VFs remain underreported on routine thoracic and abdominal CT scans. Consequently, many patients may not receive appropriate osteoporosis evaluation and treatment, placing them at increased risk of subsequent fractures, disability, and increased mortality.