Automatic bolus tracking in abdominal CT scans with convolutional neural networks

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Bolus tracking optimizes scan timing in contrast-enhanced CT but is prone to manual errors.
• Operator variability impacts enhancement levels and diagnostic accuracy.
• Current methods are time-consuming and lack standardization.

Purpose of the Study:

• To develop an AI-driven solution for fully automated bolus tracking in abdominal CT.
• To improve standardization and diagnostic accuracy.
• To simplify the imaging workflow for contrast-enhanced CT exams.

Main Methods:

• Retrospective study using diverse abdominal CT exams.
• Two-step AI approach: automatic locator scan positioning and automatic region-of-interest (ROI) positioning.
• Locator scan positioning used transfer learning for regression; ROI positioning used segmentation.

Main Results:

• Locator scan positioning network demonstrated improved consistency over manual methods.
• Locator scan positioning achieved sub-centimeter error (9.76±6.78 mm) on test data.
• ROI segmentation network achieved sub-millimeter error (0.99±0.66 mm) on test data.

Conclusions:

• AI-based locator scan positioning reduces inter-operator variance.
• Automated bolus tracking standardizes and simplifies contrast-enhanced CT workflows.
• The AI method minimizes operator-dependent decisions, enhancing reliability.