Medical Image Processing for CT Data Truncation Artifacts

The patent creates a virtual copy of the missing projection data by training a neural network model on complete detector data and using it to generate synthetic projection data that replicates what a complete detector would have captured. This virtual copying allows the system to achieve complete detector performance using only an incomplete physical detector, resolving the contradiction between image accuracy and detector cost.

The patent converts the harmful effect of data truncation from an incomplete detector into a beneficial training opportunity. By intentionally creating truncated data scenarios during model training and using them to teach the neural network how to reconstruct complete data, the system transforms the limitation of incomplete detectors into a strength, enabling high-quality images despite using fewer detector elements.

The patent performs preliminary action by pre-training the neural network model offline using complete projection data before actual medical imaging. This advance preparation stores the recovered relationships and patterns in the trained model, so that during actual scanning with incomplete detectors, the complex recovery process has already been accomplished, requiring only straightforward inference rather than complex real-time processing.