Effectiveness of Deep Learning Algorithms to Determine Laterality in Radiographs

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Radiograph laterality classification is crucial for accurate diagnosis.
• Manual classification is prone to human error, impacting patient safety.
• Automating laterality classification can improve efficiency and reduce errors.

Purpose of the Study:

• To develop and validate a highly accurate deep learning model for radiograph laterality classification.
• To compare the model's performance against human performance.
• To assess the model's utility in a rigorous, automated classification use-case.

Main Methods:

• Retrospective extraction of Digital Imaging and Communications in Medicine (DICOM) data for nine body parts.
• Development of classification and object detection deep learning models.
• Data augmentation for class balancing and rigorous testing on novel images.
• Evaluation using Receiver Operating Characteristic (ROC) curves, accuracy, sensitivity, and specificity.

Main Results:

• The final classification model achieved an Area Under the Curve (AUC) of 0.999 for novel images and per-study analysis.
• Object detection models demonstrated over 99% accuracy at both image and study levels.
• The ensemble model showed performance comparable or superior to human experts in automated classification.

Conclusions:

• Deep learning models can classify radiograph laterality with high accuracy and reliability.
• This technology has the potential to enhance patient safety and radiologist satisfaction.
• Automated classification is achievable for high-specificity use-cases in medical imaging.