Assessment of vascular invasion of pancreatic ductal adenocarcinoma based on CE-boost black blood CT technique

Yue Lin, Tongxi Liu, Yingying Hu, Yinghao Xu, Jian Wang, Sijia Guo, Sheng Xie, Hongliang SunInsights 2024 · DOI

TL;DR

This paper demonstrates that combining Deep Learning Reconstruction (DLR) with a "Black Blood" CT (BBCT) subtraction technique significantly improves the detection of vascular invasion in pancreatic cancer. By reducing image noise and enhancing the visualization of the vessel wall, this approach achieves a high diagnostic accuracy (AUC of 0.942), providing surgeons with a more reliable roadmap for determining if a tumor is resectable.

What Problem Does This Paper Address?

Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal malignancies, and surgical resection is often the only path to a cure. However, surgery is only viable if the tumor has not invaded major abdominal blood vessels (like the celiac artery or portal vein).

Currently, Computed Tomography (CT) is the gold standard for staging, but it has a "bright blood" problem. In standard contrast-enhanced CT, the blood inside the vessel is so bright that it can obscure the actual vessel wall, making it difficult to tell if a tumor is just touching the vessel or actually invading it. Furthermore, standard image reconstruction methods often struggle with "noise," which can blur the fine details needed for accurate staging.

Methods

The researchers evaluated 35 patients using a 320-row CT scanner and compared four different image processing strategies:

  1. AIDR 3D: The standard hybrid-iterative reconstruction method used to reduce noise.
  2. AiCE (Advanced intelligent clear-IQ engine): A newer Deep Learning Reconstruction (DLR) tool that uses neural networks to distinguish between true anatomical signals and noise.
  3. AIDR 3D–BBCT: The standard method combined with the Black Blood technique.
  4. AiCE–BBCT: The deep learning method combined with the Black Blood technique.

The "Black Blood" (BBCT) Trick: This is a post-processing subtraction technique. By subtracting the arterial phase image from the delayed phase image, the researchers effectively "turn off" the bright signal from the blood inside the vessel (the lumen). This highlights the vessel wall itself, making the interface between the tumor and the vessel much clearer.

Datasets

The study utilized a prospective dataset of 35 consecutive patients diagnosed with nonmetastatic PDAC at a single hospital between January 2023 and February 2024.

  • Vessels Evaluated: A total of 175 vessels were initially screened, with 30 vessels in 12 patients confirmed to have invasion findings.
  • Gold Standard: The "ground truth" for vascular invasion was confirmed via intraoperative records (surgical findings) and final pathology reports.

Key Insights and Ideas

  • Synergy of AI and Subtraction: While AiCE is excellent at cleaning up noise, it doesn't change the physics of contrast. BBCT changes the visualization by darkening the blood. Together, they create a high-fidelity image where the vessel wall "pops" against the background.
  • Superior Consistency: One of the biggest hurdles in radiology is "inter-observer variability"—two doctors looking at the same scan and seeing different things. The AiCE–BBCT group showed the highest consistency (ICC: 0.847–0.935), suggesting this method makes the diagnosis more objective.
  • DLR Outperforms Hybrid-IR: The study reinforces that deep learning-based reconstruction (AiCE) provides significantly better image quality and lower noise than traditional iterative methods (AIDR 3D) without requiring a higher radiation dose.

Key Results

The AiCE–BBCT combination was the clear winner across all metrics:

  • Diagnostic Accuracy: Achieved an Area Under the Curve (AUC) of 0.942, significantly higher than standard methods.
  • Sensitivity and Specificity: It reached a sensitivity of 93.3% and a specificity of 95.0% for detecting vascular invasion.
  • Image Quality: Both AiCE and AiCE–BBCT groups showed superior image noise reduction and diagnostic acceptability compared to the AIDR 3D groups (p < 0.001).
  • Clinical Impact: The improved visualization allowed for better assessment of "encasement" (tumor contact > 180° of the vessel) versus "abutment" (≤ 180°), which is the critical threshold for surgical decision-making.

Limitations and Future Work

  • Sample Size: The study was limited to 35 patients at a single center. Larger, multi-center trials are needed to validate these findings across broader populations.
  • Motion Artifacts: Because BBCT relies on subtracting images taken at different times, any patient movement between phases can cause "ghosting" or blurred artifacts.
  • Low-Dose Potential: The authors suggest that because AiCE is so good at denoising, future research should investigate whether they can lower the radiation dose even further while maintaining these high-accuracy results.
  • Complexity: The subtraction process adds a layer of post-processing that needs to be streamlined for high-volume clinical workflows.

Summary updated

Abstract

Identifying vascular invasion of PDAC is important for prognostication. Combined images provide improved image quality and higher diagnostic accuracy. Combined images can excellently display the vascular wall and invasion.
BibTeX
@inproceedings{lin2024assessment,
  author = {Yue Lin and Tongxi Liu and Yingying Hu and Yinghao Xu and Jian Wang and Sijia Guo and Sheng Xie and Hongliang Sun},
  title = {{Assessment of vascular invasion of pancreatic ductal adenocarcinoma based on CE-boost black blood CT technique}},
  booktitle = {Insights},
  year = {2024},
  doi = {10.1186/s13244-024-01870-x},
}
Figure