CT slice alignment to whole-body reference geometry by convolutional neural network.

Area of Science:

• Medical Imaging
• Anatomical Coordinate Systems
• Machine Learning in Radiology

Background:

• Volumetric medical imaging lacks a standardized coordinate system, hindering precise anatomical localization and advanced image analysis.
• Current methods rely on visual assessment, limiting automated image processing and data integration.

Purpose of the Study:

• To introduce a patient-size-scaled geometric system for standardizing anatomical coordinates in medical imaging.
• To develop and validate a machine learning model for predicting anatomical location from computed tomography (CT) images.

Main Methods:

• A novel geometric system was designed and applied to CT imaging data.
• A convolutional neural network (CNN) was trained to correlate axial CT slice appearance with superior-inferior anatomical position.
• The model's accuracy was evaluated across diverse anatomical regions.

Main Results:

• The trained CNN achieved an accuracy of ±12 mm in predicting per-slice reference locations.
• The model demonstrated consistent performance across various annotated regions, from the brain to the thighs.
• A functional model and training scripts were made publicly available.

Conclusions:

• The proposed geometric system and CNN provide a standardized, machine-readable coordinate system for volumetric medical imaging.
• This approach facilitates automated anatomical localization, improving tasks such as organ identification and image registration.
• The findings pave the way for enhanced medical image analysis and quality control applications.