Estimation of patient's angle from skull radiographs using deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiography

Background:

• Skull radiography is crucial for diagnosis and follow-up but often requires repeat imaging due to positioning errors.
• Accurate patient positioning in skull radiography demands significant expertise from radiologic technologists.
• Current methods for assessing patient angle in skull radiography are experience-dependent and can be time-consuming.

Purpose of the Study:

• To develop and validate a novel deep learning model for automated patient angle estimation from skull radiographs.
• To improve the accuracy and efficiency of skull radiography by reducing the need for retakes.
• To investigate the potential of artificial intelligence in optimizing radiographic procedures.

Main Methods:

• A deep learning model, specifically a residual neural network with modifications (Parametric ReLU, dropout), was developed.
• Skull radiographs were simulated using 2D projections from head CT images for supervised training.
• The model was trained to estimate patient angles in both lateral and superior-inferior directions.

Main Results:

• The deep learning model achieved high accuracy in angle estimation.
• Estimation errors were 0.56±0.36° for the lateral angle and 0.72±0.52° for the superior-inferior angle.
• The model demonstrated the feasibility of automated angle assessment in skull radiography.

Conclusions:

• Deep learning models can accurately estimate patient angles from radiographs, significantly reducing retake rates.
• This AI-driven approach can streamline skull radiography workflows and improve diagnostic efficiency.
• The developed model shows promise for facilitating more consistent and accurate skull radiography practices.