Deep learning-based Hounsfield unit value measurement method for bolus tracking images in cerebral computed...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Patient movement during bolus tracking (BT) significantly impacts Hounsfield unit (HU) measurement accuracy.
• Accurate HU values in the internal carotid artery (ICA) are crucial for diagnosis and treatment planning.

Purpose of the Study:

• To assess the accuracy of a novel deep learning (DL)-based method for measuring HU values in the ICA.
• To compare the DL method's performance against the conventional region of interest (ROI) setting method, particularly in the presence of patient movement.

Main Methods:

• Retrospective analysis of 722 BT images from 127 patients undergoing cerebral computed tomography angiography.
• Development and application of a convolutional neural network for ICA segmentation using DL.
• Comparison of HU values obtained via DL method and ROI setting (corrected and settled) to evaluate accuracy differences.

Main Results:

• The DL-based method showed significantly smaller HU value differences from the corrected ROI compared to the settled ROI method.
• In patients with body movement, differences were 9.0 ± 6.4 HU (DL) vs. 23.8 ± 12.7 HU (settled ROI).
• In patients without body movement, differences were 3.9 ± 4.7 HU (DL) vs. 1.1 ± 1.6 HU (settled ROI).

Conclusions:

• The DL-based method demonstrates superior accuracy in HU value measurements for the ICA in BT images.
• This approach effectively mitigates the inaccuracies caused by involuntary patient movement during image acquisition.