AI ANALYSIS REPORT
Developing an artificial intelligence tool for detecting fractures of child abuse: preliminary findings
This pilot study demonstrates promising results from retraining a deep learning AI model (BoneView) to detect inflicted fractures in children. Initial performance for detecting child abuse-related fractures was low, but after preliminary retraining with a dataset of 329 annotated positive studies, sensitivity improved from 44% to 52% and specificity from 61% to 67%. The findings suggest that further annotation and training could lead to clinically acceptable performance, offering an additional diagnostic opinion for complex cases like child abuse skeletal surveys, which often require double reporting by specialist radiologists.
Executive Impact
Leveraging AI for medical image analysis can deliver significant benefits across key performance indicators. Here’s how enhanced detection capabilities in paediatric radiology could translate into tangible improvements for your enterprise.
0
Baseline Sensitivity
0
Retrained Sensitivity
0
Baseline Specificity
0
Retrained Specificity
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
6.9%
of children in the UK suffer physical abuse, with fractures in up to 55% of cases.
AI Model Retraining Process
Initial BoneView Evaluation
→
Dataset Collection & Ground Truth Definition
→
Radiograph Annotation
→
Data Partition (8:2 training/test)
→
Model Retraining (Transfer Learning)
→
Performance Assessment
| Metric | Baseline Model (95% CI) | Retrained Model (95% CI) |
|---|---|---|
| Sensitivity | 44% (35%, 58%) | 52% (43%, 64%) |
| Specificity | 61% (52%, 71%) | 67% (58%, 78%) |
| AUC | 0.46 (0.38, 0.57) | 0.55 (0.48, 0.66) |
| Note: Sensitivity and specificity are calculated at an operating point of 0.5 false positives per study. Evaluation metrics are calculated at the study level, derived from the ROC curve. Values in parentheses are 95% confidence intervals, computed via bootstrapping. AUC area under the receiver operating characteristic curve. | ||
Addressing Gaps in Paediatric Radiology
The Royal College of Radiologists recommends that skeletal surveys for suspected physical abuse (SPA) be reviewed within 24 hours by two specialist paediatric radiologists. However, a UK survey indicated that only 52% of departments have a paediatric radiologist, and an audit found that only 75% of skeletal surveys met the 24-hour reporting timeframe. An AI tool could serve as a 'first reviewer' to help fill this critical gap, ensuring timely and accurate detection of subtle abuse-related fractures and improving patient outcomes.
Calculate Your Potential ROI
Timely and accurate detection of child abuse fractures can significantly reduce recurrence rates and mortality, potentially saving lives and healthcare costs associated with long-term care and repeated investigations.
Annual Savings
$0
Hours Reclaimed Annually
0
Your AI Implementation Roadmap
Our structured approach ensures a smooth and successful integration of AI, maximizing your returns and minimizing disruption.
Data Acquisition & Annotation Completion
Complete the full annotation of the dataset, aiming for a larger and more diverse set of inflicted fracture images across multiple anatomical locations.
Advanced Model Training & Optimization
Further train the deep learning model using the expanded dataset, incorporating sub-analysis by fracture type, age, displacement, and location to refine detection accuracy.
Multi-Centre Validation & Clinical Integration
Conduct external validation in a multi-centre setting to ensure reproducibility and clinical applicability, then explore integration into existing PACS workflows for an additional diagnostic opinion.
Ready to Transform Your Enterprise with AI?
Book a personalized strategy session with our AI experts to discuss how these insights can be tailored to your specific organizational needs and drive measurable results.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat