Development of retake support system for lateral knee radiographs by using deep convolutional neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Radiography Quality Control

Background:

• High retake rates in lateral knee radiography are often due to positioning errors.
• Developing automated systems can improve efficiency and reduce patient burden.

Purpose of the Study:

• To develop and evaluate a system for classifying tilting directions in lateral knee radiographs.
• To reduce the number and time of image retakes in knee radiography.

Main Methods:

• A deep convolutional neural network (DCNN) was trained using synthetic Raysum images generated from 3D CT data.
• The system classified tilting into four directional categories.
• Testing involved Raysum images, phantom images, and rejected clinical radiographs.

Main Results:

• The DCNN achieved overall accuracies of 88.5%, 81.4%, and 73.3% on different test datasets.
• Classification accuracy increased with larger tilting degrees.
• The use of synthetic Raysum images facilitated dataset creation for DCNN training.

Conclusions:

• Deep convolutional neural networks can effectively classify knee joint tilting from lateral radiographs.
• Synthetic imaging data aids in creating datasets for training AI models.
• This system shows potential as a support tool for lateral knee radiography, improving efficiency and patient care.