Deep Learning for Standardized Head CT Reformatting: A Quantitative Analysis of Image Quality and Operator...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Manual head CT reformatting is crucial for diagnosis but can be time-consuming and variable.
• Assessing the quality and efficiency of manual reformats is essential for optimizing clinical workflows.

Purpose of the Study:

• To validate a deep learning foundation model for automated head CT reformatting.
• To quantify the quality, speed, and variability of conventional manual reformats.

Main Methods:

• An AI foundation model generated automated reformats for 1,763 head CT scans.
• Model accuracy was validated against expert manual annotations for landmark detection and reformat errors.
• The AI model served as a reference to evaluate technician-generated reformats.

Main Results:

• AI model showed high concordance with expert annotations (landmark error 0.6-0.9 mm).
• AI reformats had minimal errors (rotational 0.7°, centering 0.3%, zoom 0.4%).
• Manual reformats exhibited significant errors (rotational 11.2°, centering 6.4%, zoom 6.2%) and variability.

Conclusions:

• Manual head CT reformatting demonstrates substantial variability in quality and speed.
• A deep learning foundation model achieves accurate and consistent automated reformats.
• AI-driven automated reformatting can enhance standardization, efficiency, and cost-effectiveness in clinical practice.