Deep Learning-Based Estimation of Radiographic Position to Automatically Set Up the X-Ray Prime Factors

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Medical Physics

Background:

• X-ray prime factors (KVp, mAs, distance) critically affect radiation dose and image quality.
• Incorrect parameter settings lead to exposure errors, necessitating repeat scans and increasing patient radiation exposure.
• Accurate radiographic positioning is essential for selecting optimal X-ray parameters.

Purpose of the Study:

• To develop and validate a deep learning model for automatic radiographic position estimation from pre-exposure photographs.
• To leverage estimated positions for optimal selection of X-ray prime factors (KVp, mAs, distance).
• To reduce exposure errors and minimize patient radiation dose in diagnostic imaging.

Main Methods:

• A novel deep learning approach using a lightweight ConvNeXt architecture was employed.
• The model was trained on a database of 66 common radiographic positions from 75 volunteers.
• Fine-tuning, discriminative learning rates, and a one-cycle policy scheduler were utilized for model optimization.

Main Results:

• The model achieved 93.17% accuracy in classifying radiographic positions from photographs.
• Accuracy in selecting correct prime factors increased to 95.58%, accounting for positions with similar parameters.
• Most classification errors occurred for positions with similar patient poses in the images.

Conclusions:

• The proposed deep learning method is feasible for automating radiographic position estimation.
• This approach can streamline the X-ray acquisition workflow and reduce exposure errors.
• The system has the potential to significantly decrease unnecessary radiation doses to patients.