BPPR: A Framework for Content Navigation in Multi-Contrast Body CT Images using Deep Regression Models

Area of Science:

• Medical Imaging
• Radiology
• Computer Vision

Background:

• Clinical computed tomography (CT) images vary significantly due to diverse scanners and acquisition parameters.
• Automated content navigation is crucial for curating CT image datasets and enabling downstream analyses.

Purpose of the Study:

• To develop an automated framework, Body-Part-Phase Regression (BPPR), for identifying regions of interest and contrast enhancement phases in body CT images.
• To enhance the navigation and analysis of diverse CT imaging data.

Main Methods:

• A two-phase body part regression method for predicting 2D slice locations within 3D volumes.
• A circular regression model for continuous prediction of CT image contrast timing relative to contrast agent injection.
• A positional weighting mechanism linking slice-level and volume-level predictions for enhanced phase prediction.

Main Results:

• The BPPR framework demonstrated improved slice localization and contrast phase prediction accuracy on large-scale multi-contrast CT datasets.
• The two-phase training scheme reduced slice localization error from 9.2 mm to 6.1 mm compared to previous methods.
• BPPR offers advantages over segmentation-based approaches for CT image content navigation.

Conclusions:

• The unified Body-Part-Phase Regression (BPPR) framework provides a cohesive approach to continuous content navigation in CT images.
• BPPR enhances the precision of anatomical localization and contrast phase determination.
• The framework has potential clinical applications for improving the utility of CT imaging data.