Detection of Traumatic Pediatric Elbow Joint Effusion Using a Deep Convolutional Neural Network

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pediatric Radiology

Background:

• Elbow effusions in pediatric patients often indicate trauma.
• Accurate and timely diagnosis is crucial for appropriate management.
• Radiographic interpretation can be challenging, especially with subtle effusions.

Purpose of the Study:

• To evaluate the diagnostic performance of a deep convolutional neural network (DCNN) for identifying traumatic pediatric elbow effusions.
• To assess if a DCNN trained on a limited dataset can achieve high accuracy.

Main Methods:

• A dataset of 901 pediatric lateral elbow radiographs was curated.
• Radiographs were divided into training (657), validation (115), and test (129) sets.
• Various DCNN architectures were trained and optimized using the validation set.

Main Results:

• The best-performing DCNN achieved an ROC AUC of 0.943 on the independent test set.
• The model demonstrated high diagnostic performance with 90.9% sensitivity and 90.6% specificity.
• Overall accuracy for diagnosing elbow effusions was 90.7% on the test set.

Conclusions:

• Deep convolutional neural networks are effective tools for diagnosing traumatic pediatric elbow effusions.
• DCNNs can achieve high accuracy even when trained on relatively limited datasets.
• This technology holds promise for improving diagnostic efficiency in pediatric radiology.