Related Experiment Video
Updated: Jun 30, 2025
09:34
A Method to Estimate Cadaveric Femur Cortical Strains During Fracture Testing Using Digital Image Correlation
Published on: September 14, 2017
7.4K
Development of a diagnostic support system for distal humerus fracture using artificial intelligence
Aashay Kekatpure1, Aditya Kekatpure2, Sanjay Deshpande3
1Department of Orthopaedics, DMMC, Nagpur, India. dr.aashayk@gmail.com.
International Orthopaedics
|March 19, 2024
Summary
This study developed an artificial intelligence (AI) system for detecting distal humerus fractures on X-rays. While specific, the AI showed high accuracy in identifying fractures but needs improved sensitivity for clinical use.
Area of Science:
- Orthopaedic Surgery
- Radiology
- Artificial Intelligence
- Deep Learning
- Medical Image Analysis
Background:
- Distal humerus fractures are clinically significant, requiring prompt diagnosis to prevent complications.
- Conventional X-ray imaging can miss subtle fractures, leading to diagnostic delays.
- Artificial intelligence (AI), particularly deep learning, offers potential for automating medical image analysis and improving diagnostic accuracy.
Purpose of the Study:
- To develop and evaluate a deep learning-based diagnostic support system for identifying distal humerus fractures.
- To assess the reliability of AI in providing image-based fracture detection recommendations using conventional X-ray images.
Main Methods:
- Retrospective analysis of 4931 conventional radiography elbow images (anteroposterior and lateral views) from patients aged seven years and older.
- Images were reviewed and labelled by two senior orthopaedic surgeons; the dataset was divided into training (79.88%) and validation (20.1%) sets.
- A ResNet18 deep learning architecture was employed after pre-processing images to 224x224 pixels.
Main Results:
- The deep learning model achieved 69.14% accuracy, 95.89% specificity, and 99.47% positive predictive value (PPV) in the validation set.
- Sensitivity was 61.49%, indicating a notable rate of false negatives.
- Area Under the Curve (AUC) was 0.787 for the AI model versus 0.580 for the senior surgeon, suggesting AI's potential but with performance variations compared to human experts.
Conclusions:
- The deep learning system demonstrates potential for diagnosing distal humerus fractures from radiographs, with strong specificity and PPV for identifying subtle lesions.
- The model's high false negative rate (low sensitivity) requires improvement for clinical application.
- Further research and validation are essential to enhance sensitivity and overall diagnostic accuracy for practical clinical implementation.
