Radiomics-Based Machine Learning Classification Strategy for Characterization of Hepatocellular Carcinoma on Contrast-Enhanced Ultrasound in High-Risk Patients with LI-RADS Category M Nodules

Lingling Li, Xiaoxin Liang, Yiwen Yu, Rushuang Mao, Jing Han, Chuan Peng, Jianhua ZhouIndian 2024 · DOI

Abstract

Objective: Accurate differentiation within the LI-RADS category M (LR-M) between hepatocellular carcinoma (HCC) and non-HCC malignancies (mainly intrahepatic cholangiocarcinoma [CCA] and combined hepatocellular and cholangiocarcinoma [cHCC-CCA]) is an area of active investigation. We aimed to use a radiomics-based machine learning classification strategy for differentiating HCC from CCA and cHCC-CCA on contrast-enhanced ultrasound (CEUS) images in high-risk patients with LR-M nodules.

Methods: A total of 159 high-risk patients with LR-M nodules (69 HCC and 90 CCA/cHCC-CCA) were retrospectively included (111 for training, 48 for testing). Six CEUS images per observation were collected across predetermined time points for tumor segmentation and selection of 7,896 radiomics features. Clinical data and key radiomics features were used to develop a clinical model, a radiomics signature (RS), and a combined RS-clinical (RS-C) model.

Results: The RS and RS-C models performed significantly better than the clinical model in the test set (AUC of 0.698 for clinical, 0.903 for RS, and 0.912 for RS-C).

Conclusions: Radiomics-based machine learning classifiers are competent for differentiating HCC from other malignancies in high-risk patients with LR-M nodules.

Introduction

The Liver Imaging Reporting and Data System (LI-RADS) provides a standardized framework for classifying liver nodules in patients at high risk for hepatocellular carcinoma (HCC). While the LR-5 category is highly specific for HCC, the LR-M category represents a diagnostic challenge. It indicates that a lesion is definitely malignant but lacks the specific imaging hallmarks of HCC, often encompassing intrahepatic cholangiocarcinoma (CCA) or combined tumors (cHCC-CCA).

Distinguishing between these is critical for treatment planning; HCC may qualify for liver transplantation, whereas CCA typically requires radical resection. Currently, many LR-M nodules require invasive biopsies. This study investigates whether "radiomics"—the quantitative extraction of features from medical imaging—can provide a non-invasive diagnostic alternative.

Methods and Materials

Patient Cohort and Dataset

The study utilized a retrospective dataset from the Sun Yat-Sen University Cancer Center, spanning January 2006 to December 2019. From an initial screening of 4,180 patients, 159 high-risk patients with LR-M nodules were selected based on strict inclusion criteria, including pathologic confirmation and CEUS imaging within 30 days of diagnosis.

Figure 1: Flowchart of the patient selection process, detailing the exclusion criteria and the final distribution of the training and test sets.

Radiomics Pipeline

The researchers developed a multi-temporal machine learning pipeline:

  1. Multi-Temporal Analysis: Images were collected at six time points, from peak enhancement (~20s) to late-phase washout (3 minutes).
  2. Feature Extraction: 7,896 radiomics features were extracted per observation, covering intensity, shape, and complex texture matrices.
  3. Dimensionality Reduction: To prevent overfitting, features were filtered using Intraclass Correlation Coefficients (ICC), minimum Redundancy Maximum Relevance (mRMR), and Least Absolute Shrinkage and Selection Operator (LASSO) regression.
  4. Model Construction: Three models were compared: a Clinical Model (using biomarkers like AFP and CA19-9), a Radiomics Signature (RS), and a combined RS-Clinical (RS-C) Model.
Figure 4: Overview of the radiomics workflow, including tumor segmentation (ROI), feature extraction, dimension reduction via LASSO, and statistical performance metrics (ROC curves, calibration plots, and decision curve analysis).

Results

Model Performance Comparison

The machine learning models drastically outperformed the standard clinical approach in the independent test set. Texture-based features, particularly those from the Gray Level Run Length Matrix, were found to be the most diagnostic, reflecting internal tumor heterogeneity and micro-vascularization.

Model AUC (95% CI) Accuracy Sensitivity Specificity
Clinical Model 0.698 (0.571–0.812) 68.8% 71.4% 66.7%
Radiomics (RS) 0.903 (0.830–0.970) 87.5% 90.5% 85.2%
Combined (RS-C) 0.912 (0.838–0.977) 85.4% 95.2% 77.8%

The combined RS-C model achieved a high sensitivity of 95.2% for HCC. Decision Curve Analysis (DCA) further confirmed that radiomics-based models provided a higher net benefit across all threshold probabilities compared to the clinical model alone.

Discussion and Conclusion

This study demonstrates that radiomics-based machine learning can significantly improve the characterization of LR-M nodules on CEUS. By capturing the temporal flow of contrast agents and the spatial arrangement of pixel intensities, these models offer a robust, non-invasive method for identifying HCC.

Limitations include the manual nature of tumor segmentation and the single-center retrospective design. Future work should focus on automated AI segmentation and external validation across multiple institutions to ensure the generalizability of these findings in clinical practice.

Summary updated

Abstract

<b>Objective</b> Accurate differentiation within the LI-RADS category M (LR-M) between hepatocellular carcinoma (HCC) and non-HCC malignancies (mainly intrahepatic cholangiocarcinoma [CCA] and combined hepatocellular and cholangiocarcinoma [cHCC-CCA]) is an area of active investigation. We aimed to use radiomics-based machine learning classification strategy for differentiating HCC from CCA and cHCC-CCA on contrast-enhanced ultrasound (CEUS) images in high-risk patients with LR-M nodules. <b>Methods</b> A total of 159 high-risk patients with LR-M nodules (69 HCC and 90 CCA/cHCC-CCA) who underwent CEUS within 1 month before pathologic confirmation from January 2006 to December 2019 were retrospectively included (111 patients for training set and 48 for test set). The training set was used to build models, while the test set was used to compare models. For each observation, six CEUS images captured at predetermined time points (T1, peak enhancement after contrast injection; T2, 30 seconds; T3, 45 seconds; T4, 60 seconds; T5, 1-2 minutes; and T6, 2-3 minutes) were collected for tumor segmentation and selection of radiomics features, which included seven types of features: first-order statistics, shape (2D), gray-level co-occurrence matrix, gray-level size zone matrix, gray-level run length matrix, neighboring gray tone difference matrix, and gray-level dependence matrix. Clinical data and key radiomics features were employed to develop the clinical model, radiomics signature (RS), and combined RS-clinical (RS-C) model. The RS and RS-C model were built using the machine learning framework. The diagnostic performance of these three models was calculated and compared. <b>Results</b> Alpha-fetoprotein (AFP), CA19-9, enhancement pattern, and time of washout were included as independent factors for clinical model (all <i>p</i> < 0.05). Both the RS and RS-C model performed better than the clinical model in the test set (area under the curve [AUC] of 0.698 [0.571-0.812] for clinical model, 0.903 [0.830-0.970] for RS, and 0.912 [0.838-0.977] for the RS-C model; both <i>p</i> < 0.05). <b>Conclusions</b> Radiomics-based machine learning classifiers may be competent for differentiating HCC from CCA and cHCC-CCA in high-risk patients with LR-M nodules.
BibTeX
@inproceedings{li2024radiomics,
  author = {Lingling Li and Xiaoxin Liang and Yiwen Yu and Rushuang Mao and Jing Han and Chuan Peng and Jianhua Zhou},
  title = {{Radiomics-Based Machine Learning Classification Strategy for Characterization of Hepatocellular Carcinoma on Contrast-Enhanced Ultrasound in High-Risk Patients with LI-RADS Category M Nodules}},
  booktitle = {Indian},
  year = {2024},
  doi = {10.1055/s-0043-1777993},
}
Figure