20 May 2026 in The Ultrasound Journal
AHAbhilash Hareendranathan
Background: Bone fractures are common in acute care, and point-of-care ultrasound (POCUS) is an emerging diagnostic tool that can be complementary to or even in some cases an alternative to X-ray imaging. With the rise of artificial intelligence (AI) model incorporation in diagnostic interpretations, this systematic review and meta-analysis aimed to evaluate the diagnostic performance of deep learning models for detecting bone fractures on ultrasound images, using radiographic imaging with expert interpretation as the reference standard. Methods: Comprehensive literature searches were conducted from inception to 15 August 2025, in databases including MEDLINE (Ovid interface), Embase (Ovid interface), CINAHL Plus with Full Text (EBSCOhost interface), Web of Science, ACM Digital Library, Scopus, and Google Scholar. We included studies evaluating deep learning models applied to ultrasound images or sweeps for fracture classification. The included studies predominantly evaluated pediatric cohorts presenting to emergency departments with suspected upper extremity fractures. Sensitivity and specificity were pooled using a bivariate random-effects model on the logit scale, and summary receiver operating characteristic (sROC) curve was constructed. Risk of bias was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool. Results: A total of 580 papers were identified in the preliminary literature search, and 333 studies were screened after duplicate removal. Screening 27 full-text articles resulted in eight studies that were included in the review. Across 21 reported model evaluations from five studies with extractable diagnostic data, pooled sensitivity was 0.77 (95% CI: 0.69-0.83) and pooled specificity was 0.90 (95% CI: 0.86-0.94), with an overall sROC AUC of 0.91. Patient-level evaluations yielded higher pooled sensitivity (0.89; 95% CI: 0.81-0.94) and specificity (0.94; 95% CI: 0.81-0.98) compared to video-level analyses. Conclusions: AI-assisted ultrasound demonstrates promising diagnostic performance for fracture detection with generally high specificity and variable sensitivity across architectures and anatomic targets. Future work should prioritize prospective, externally validated studies with standardized acquisition protocols and clinically meaningful reporting at the patient level.
1 October 2025 in Radiology
3 June 2026 in Computers in Biology and Medicine
It is often desirable to generalize medical imaging AI models trained with dense annotations to data acquired from different ultrasound scanners or clinical sites; however, retraining these models with new annotations is often difficult and costly. We examine this challenge in pediatric wrist fracture assessment using point-of-care ultrasound (POCUS), where fractures are common and can be effectively triaged via ultrasound. AI has shown radiologist-level performance for fracture detection, often aided by high-quality bony structure segmentation. However, due to significant domain shifts, models perform poorly on data from other centers or probes, and obtaining segmentation labels across devices is impractical due to manual annotation effort and data privacy concerns. To address this, we propose a target-informed self-supervised pretraining and model-ensemble strategy. Specifically, our approach combines masked image modeling (MIM) and contrastive learning to learn target-domain structural representations without labels, and introduces a confidence-aware infusion head to adaptively integrate predictions. The source dataset, collected with a Philips Lumify probe, contained dense labels, while the target dataset, acquired with a TeleMED portable probe, was unlabeled. The datasets were kept strictly separate throughout the entire process. Our method used labeled source data for supervised training and leveraged target-domain pretraining to improve generalization. On 318 images from 62 pediatric POCUS videos, this approach significantly improved cross-device performance, achieving over 6% Dice improvement on the target domain versus the baseline. These results demonstrate a label-efficient and privacy-preserving approach for cross-device-robust ultrasound AI, offering a framework that can be extended to multi-center studies or federated learning setups. GitHub Repository:https://github.com/yuyue2uofa/CrossDomainPOCUS.