Comparative analysis of two mathematical algorithms for the calculation of computed tomography perfusion parameters in the healthy and diseased pancreas

Kehua Pan, Hongqing Wang, Xiaoyu Chen, Ye Xiaocui, Zhao Zhang, Xiufen JiaJournal of Applied Clinical Medical Physics 2021 · DOI

TL;DR

This study compares two common algorithms, Maximum Slope (MS) and Deconvolution (DC), for calculating pancreatic blood flow (BF) and blood volume (BV) from CT perfusion (CTP) data. While both algorithms show a good correlation, they produce significantly different quantitative values, meaning results from one cannot be directly interchanged with the other in clinical practice.

What Problem Does This Paper Address?

Computed Tomography Perfusion (CTP) imaging is a valuable tool for diagnosing and monitoring pancreatic diseases. However, the quantitative parameters derived from CTP data can vary depending on the post-processing algorithm used. Two prevalent algorithms are the Maximum Slope (MS) and Deconvolution (DC) methods. Clinicians and researchers need to understand if these algorithms produce comparable results, especially when interpreting CTP data for diagnosis, treatment monitoring, and clinical trials. This paper addresses the critical question of how the MS and DC algorithms compare in their calculation of pancreatic perfusion parameters (blood flow and blood volume) in both healthy individuals and patients with acute pancreatitis, highlighting the potential impact of algorithm choice on clinical interpretation.

Methods

The researchers retrospectively analyzed CTP data from 57 patients (30 healthy, 27 with acute pancreatitis) using two distinct software platforms: one employing the MS algorithm and another using the DC algorithm. Two senior radiologists, blinded to the clinical data and algorithm type, independently analyzed the images. Regions of Interest (ROIs) were placed in the head, body, and tail of the pancreas.

Key analytical techniques included:

  • Paired t-tests: To compare the mean differences in CTP parameters (BV and BF) between the MS and DC algorithms within the same pancreatic regions.
  • Pearson linear correlation analysis: To assess the strength of the linear relationship between the parameters calculated by the two algorithms.
  • Bland-Altman analysis: To evaluate the agreement and consistency between the measurements from the MS and DC algorithms, identifying the magnitude of differences and potential biases.
  • Interobserver consistency analysis: Using Intraclass Correlation Coefficients (ICC) to ensure reliability between the two radiologists.

Datasets

The study utilized retrospective CTP data from 57 patients who underwent imaging between January and August 2015 at the First Affiliated Hospital of Wenzhou Medical University. The dataset comprised:

  • 30 subjects with a normal pancreas.
  • 27 subjects with acute interstitial pancreatitis, confirmed by clinical, laboratory, and imaging findings.

No new benchmarks were introduced; the study focused on comparing existing algorithms on a clinical dataset.

Key Insights and Ideas

  • Algorithm choice matters: The most significant takeaway is that while the MS and DC algorithms are correlated, their absolute values for pancreatic blood flow and blood volume are not interchangeable. This has direct implications for clinical decision-making.
  • MS overestimates BV and BF: The MS algorithm consistently yielded higher values for both blood volume and blood flow compared to the DC algorithm, particularly for blood volume. This difference was statistically significant in both healthy and pancreatitis groups.
  • Theoretical basis for differences: The paper suggests that the MS algorithm's assumption of no contrast agent outflow before reaching the maximum slope might lead to an overestimation of perfusion parameters, especially in physiological states where outflow is present. The DC algorithm, by contrast, is presented as being less reliant on specific perfusion models and potentially more reflective of actual physiological states.
  • Consistency is lacking: Despite good correlation, the Bland-Altman analysis revealed large absolute differences and wide 95% consistency intervals, indicating poor agreement for clinical use. This means that thresholds established using one algorithm cannot be directly applied when using the other.

Key Results

  • Blood Volume (BV): In normal pancreas subjects, BV values were markedly higher with MS (mean difference of 33.2 ml/100g) than DC. In acute pancreatitis patients, this difference was even larger (mean difference of 43.9 ml/100g). These differences were statistically significant (p < 0.001 for both groups).
  • Blood Flow (BF): BF values were also higher with MS than DC, though the difference was less pronounced than for BV. For normal pancreas subjects, the mean BF difference was 3.8 ml/min·100g (p = 0.031). For acute pancreatitis patients, the mean BF difference was 10.4 ml/min·100g (p < 0.001).
  • Correlation: Pearson linear correlation analysis showed a good correlation between BF and BV values calculated by both algorithms (r > 0.6 for normal pancreas, r > 0.7 for pancreatitis).
  • Consistency: Bland-Altman analysis indicated poor consistency, with a significant number of outside data points and large absolute differences, suggesting the values are not interchangeable.
  • Interobserver Reliability: The consistency between the two radiologists was good (ICC ≥ 0.9).

Limitations and Future Work

The authors acknowledge several limitations:

  • ROI Placement: The location and size of the ROIs could influence measurement results.
  • Scanning Initiation Time: The 10-second delay in contrast injection might affect results, particularly for the MS method, which is more sensitive to cardiac parameters.
  • Contrast Agent Flow Rate: The high flow rate of the contrast agent used might also impact the results, potentially limiting generalizability to routine clinical practice.

Future work should focus on standardizing CTP post-processing methods and establishing algorithm-specific thresholds for clinical interpretation. The study strongly suggests that when using pancreatic CTP parameters for diagnosis, severity assessment, or monitoring treatment efficacy, it is crucial to be aware of and clearly state which algorithm was used, as the results are not interchangeable.

Summary updated

Abstract

The BF and BV values measured using MS and DC algorithms had a good correlation but were not consistent.
BibTeX
@article{pan2021comparative,
  author = {Kehua Pan and Hongqing Wang and Xiaoyu Chen and Ye Xiaocui and Zhao Zhang and Xiao Chen and Xiufen Jia},
  title = {{Comparative analysis of two mathematical algorithms for the calculation of computed tomography perfusion parameters in the healthy and diseased pancreas}},
  journal = {Journal of Applied Clinical Medical Physics},
  year = {2021},
  doi = {10.1002/acm2.13488},
}
Figure