Negative Screening AI in Ankle Stress Radiography to Reduce Workload

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Interpreting ankle stress radiographs is often subjective and time-consuming.
• Developing an AI model can potentially streamline the screening of negative cases.

Purpose of the Study:

• To train an AI model for efficient screening of ankle stress radiographs.
• To assess the agreement between AI-assisted readings and expert radiologists.
• To quantify the workload reduction achieved with AI assistance.

Main Methods:

• Anterior draw test (ADT) and talar tilt test (TTT) radiographs were collected.
• A VGG16 AI model was trained and tested on independent datasets.
• Radiologists evaluated images with and without AI assistance, measuring agreement and workload.

Main Results:

• AI assistance did not significantly alter agreement with expert interpretations.
• AI alone showed fair to moderate agreement with expert readings.
• AI-assistance reduced hypothetical workload by 58.3-89.2% across different tests.

Conclusions:

• An AI model for ankle stress radiography was successfully trained.
• The AI model achieved an average workload reduction of 70% while maintaining expert agreement.
• AI shows promise in improving efficiency for ankle radiograph interpretation.