Three-Dimensional Voxel-Wise Quantitative Assessment of Imaging Features in Hepatocellular Carcinoma

Chongfei Huang, Shihong Ying, Meixiang Huang, Chenhui Qiu, Lu Fang, Zhiyi Peng, Dexing KongDiagnostics 2023 · DOI

TL;DR

This paper introduces a novel 3D voxel-wise quantitative model for assessing key imaging features (arterial phase hyperenhancement and washout) of hepatocellular carcinoma (HCC) in multiphase CT scans. By employing a robust nonlinear registration technique to overcome image misalignments, the method aims to provide more objective and accurate diagnoses compared to current subjective visual assessments.

What Problem Does This Paper Address?

Hepatocellular carcinoma (HCC) is a major cause of cancer-related deaths, and its diagnosis heavily relies on interpreting specific imaging features in CT scans, particularly arterial phase hyperenhancement (APHE) and subsequent washout (WO). Current diagnostic guidelines, like LI-RADS, involve radiologists visually assessing these features across different CT phases. However, this qualitative approach is prone to subjective interpretation, leading to inter-reader variability and potentially impacting diagnostic accuracy. Furthermore, performing voxel-wise quantitative analysis in 3D is challenging due to image misalignments caused by patient breathing and other physiological movements between scans. This paper aims to bridge this gap by developing a quantitative, 3D voxel-wise assessment method that addresses these limitations.

Methods

The core of this work is a registration-based quantitative model. The process involves several key steps:

  1. Image Preprocessing: Semi-automatic liver segmentation and expert-marked lesion areas are performed. Rigid registration is used for initial alignment.
  2. Nonlinear Registration: To handle the significant local deformations caused by respiration and other movements, the authors employ an Iterative Reweighted Local Cross-Correlation (IRLCC) method. This method is designed to capture piecewise-smooth deformations, which is crucial for dealing with the discontinuous nature of sliding motions in the liver. The IRLCC method minimizes an energy functional that balances image similarity, deformation smoothness, and image priors.
  3. Feature Location Extraction: After registration, the model identifies the locations of APHE and WO features. This is done by defining specific conditions based on intensity differences between the registered CT phases (Pre, AP, DP) and comparing them to surrounding liver tissue. This approach aims to avoid the ambiguity of setting arbitrary intensity thresholds.
  4. Voxel-Wise Quantitative Assessment: Once the feature locations are identified, the model quantifies the degree of APHE and WO for each voxel within these regions. For APHE, it calculates the percentage of arterial enhancement (PAE). For WO, it uses an improved lesion-to-liver contrast ratio (LLCR) metric.
  5. Visualization and Analysis: The quantified results are visualized as heatmaps, providing a 3D representation of feature distribution and intensity. This allows for a more comprehensive understanding of the lesion characteristics.

Datasets

The study utilized a retrospective dataset of 35 multiphase CT images from patients who underwent liver resection for HCC between 2014 and 2015. All data were pathologically proven to be HCC. The CT scans were acquired using either Toshiba or Philips scanners. The dataset included patient demographics, tumor size, and imaging parameters. No new benchmark datasets were introduced; the study used existing clinical data.

Key Insights and Ideas

  • The critical role of registration: The paper strongly emphasizes that accurate 3D quantitative assessment is impossible without robust registration. The proposed IRLCC method effectively handles the complex, piecewise-smooth deformations in multiphase liver CT, which is a significant improvement over methods assuming global smoothness.
  • Moving beyond subjective visual assessment: The core idea is to replace subjective visual interpretation with objective, quantitative metrics. This has the potential to significantly improve inter-reader agreement and diagnostic consistency.
  • Voxel-wise analysis for comprehensive feature characterization: By analyzing each voxel, the model captures the full 3D extent and intensity variations of the features, providing a more complete picture than traditional 2D region-of-interest (ROI) based methods.
  • Feature extraction based on definition, not just threshold: The method defines feature locations based on the LI-RADS criteria, rather than relying on arbitrary intensity thresholds, which can be problematic and lead to errors.

Key Results

  • Registration Accuracy: The proposed nonlinear registration framework achieved high accuracy, as evidenced by:
  • Liver: Mean Dice Similarity Coefficient (DSC) of 98.6% (Pre-AP) and 98.1% (DP-AP). Mean surface distance (MSD) of 0.38 mm and Hausdorff distance (HDD) of 4.34 mm for Pre-AP registration.
  • Lesion: Mean DSC of 98.7% (Pre-AP) and 98.3% (DP-AP). Mean MSD of 0.31 mm and HDD of 2.24 mm for Pre-AP registration. These results significantly outperformed no registration and rigid registration.

  • Feature Classification:

    • For APHE classification, the results obtained by the proposed method were consistent with those acquired by experts.
  • For WO classification, the effectiveness of the model was verified in terms of WO volume ratio, with an Area Under the Curve (AUC) of 89.8% on the ROC curve.

Limitations and Future Work

  • Segmentation Dependency: The accuracy of the proposed method relies on accurate liver segmentation. The current semi-automatic segmentation is time-consuming, and the authors plan to investigate automatic segmentation algorithms with high accuracy.
  • Post-processing for Smoothness: While the method identifies feature areas, these areas may not be perfectly smooth. Future work will involve incorporating post-processing methods to achieve smoother segmentation masks.
  • Integration into Software: The authors aim to integrate the quantitative assessment method into software to assist radiologists in diagnosing patients.
  • Further Treatment Evaluation: The quantified results could potentially be used to calculate the volume of active tumors and aid in treatment evaluation.
  • Diagnostic Accuracy Improvement: The authors acknowledge that further improvement is needed to match the diagnostic accuracy of qualitative LI-RADS.

Summary updated

Abstract

Voxel-wise quantitative assessment of typical characteristics in three-dimensional (3D) multiphase computed tomography (CT) imaging, especially arterial phase hyperenhancement (APHE) and subsequent washout (WO), is crucial for the diagnosis and therapy of hepatocellular carcinoma (HCC). However, this process is still missing in practice. Radiologists often visually estimate these features, which limit the diagnostic accuracy due to subjective interpretation and qualitative assessment. Quantitative assessment is one of the solutions to this problem. However, performing voxel-wise assessment in 3D is difficult due to the misalignments between images caused by respiratory and other physiological motions. In this paper, based on the Liver Imaging Reporting and Data System (v2018), we propose a registration-based quantitative model for the 3D voxel-wise assessment of image characteristics through multiple CT imaging phases. Specifically, we selected three phases from sequential CT imaging phases, i.e., pre-contrast phase (Pre), arterial phase (AP), delayed phase (DP), and then registered Pre and DP images to the AP image to extract and assess the major imaging characteristics. An iterative reweighted local cross-correlation was applied in the proposed registration model to construct the fidelity term for comparison of intensity features across different imaging phases, which is challenging due to their distinct intensity appearance. Experiments on clinical dataset showed that the means of dice similarity coefficient of liver were 98.6% and 98.1%, those of surface distance were 0.38 and 0.54 mm, and those of Hausdorff distance were 4.34 and 6.16 mm, indicating that quantitative estimation can be accomplished with high accuracy. For the classification of APHE, the result obtained by our method was consistent with those acquired by experts. For the WO, the effectiveness of the model was verified in terms of WO volume ratio.
BibTeX
@inproceedings{huang2023three,
  author = {Chongfei Huang and Shihong Ying and Meixiang Huang and Chenhui Qiu and Lu Fang and Zhiyi Peng and Dexing Kong},
  title = {{Three-Dimensional Voxel-Wise Quantitative Assessment of Imaging Features in Hepatocellular Carcinoma}},
  booktitle = {Diagnostics},
  year = {2023},
  doi = {10.3390/diagnostics13061170},
}
Figure