Abstract
Objectives
To develop and validate a whole-liver radiomic model using multiparametric MRI for predicting early-stage liver fibrosis (LF) in rabbits.
Methods
A total of 134 rabbits (early-stage LF, n = 91; advanced-stage LF, n = 43) who underwent liver magnetic resonance elastography (MRE), hepatobiliary phase, dynamic contrast enhanced (DCE), intravoxel incoherent motion (IVIM), diffusion kurtosis imaging, and T2* scanning were enrolled and randomly allocated to either the training or validation cohort. Whole-liver radiomic features were extracted and selected to develop a radiomic model and generate quantitative Rad-scores. Then, multivariable logistic regression was utilized to determine the Rad-scores associated with early-stage LF, and effective features were integrated to establish a combined model. The predictive performance was assessed by the area under the curve (AUC).
Results
The MRE model achieved superior AUCs of 0.95 in the training cohort and 0.86 in the validation cohort, followed by the DCE-MRI model (0.93 and 0.82), while the IVIM model had lower AUC values of 0.91 and 0.82, respectively. The Rad-scores of MRE, DCE-MRI and IVIM were identified as independent predictors associated with early-stage LF. The combined model demonstrated AUC values of 0.96 and 0.88 for predicting early-stage LF in the training and validation cohorts, respectively.
Conclusions
Our study highlights the remarkable performance of a multiparametric MRI-based radiomic model for the individualized diagnosis of early-stage LF.
Advances in knowledge
This is the first study to develop a combined model by integrating multiparametric radiomic features to improve the accuracy of LF staging.
Keywords: radiomics, multiparametric MRI, liver fibrosis, early stage, animal model
Introduction
Liver fibrosis (LF) is a progressive process characterized by excess accumulation of substances in the extracellular space of the liver, and its incidence is increasing worldwide. The alteration of hepatic architecture with fibrotic changes, along with the impairment of liver function resulting from damage to hepatocytes and the liver parenchyma by various aetiologies, contributes to the development of LF.1,2 For early-stage LF, specific antifibrotic therapies or strategies addressing the underlying cause can reverse its progression.3,4 However, once LF reaches advanced stages such as cirrhosis, it becomes irreversible and increases the risk of liver cancer, portal hypertension, and liver failure.
Liver transplantation is a life-saving option for patients with cirrhosis, but there is a shortage of donor organs compared to the demand. Therefore, early detection and accurate staging of LF are crucial to prompt intervention and prevent or delay further decline, reducing the need for liver transplantation.5 Liver biopsy has traditionally been regarded as the reference standard to confirm LF stage. However, it is associated with several drawbacks, such as potential sampling errors due to heterogeneous dust, invasive complications, and interobserver variability, making it less desirable for patients.6 Consequently, there is a compelling need for noninvasive, highly accurate, and standardized quantitative imaging techniques that are effective in diagnosing and monitoring LF progression.
Radiomics is an emerging tool used in medical imaging that quantifies texture characteristics regarding signal intensities and pixel interrelationships using mathematical methods. It overcomes the subjective nature of image interpretation and provides information that is not assessable by the naked eye. Multiparametric MRI, such as MRE, hepatobiliary phase (HBP), DCE-MRI, intravoxel incoherent motion (IVIM), diffusion kurtosis imaging (DKI), and the T2* sequence, has emerged as the most promising modality for evaluating the liver parenchyma and is acknowledged for its high accuracy in staging LF.7,8 These quantitative methods provide not only morphological information but also valuable insights into liver stiffness, vascular perfusion, permeability, the expression of hepatocyte transporters, molecular diffusion and iron quantification, which can be performed to assess not only the stage of LF but also the severity of cirrhosis.9–11
Radiomics based on various functional MRI provides a further step in the research of diffuse LF; consequently, we hypothesize that radiomics analysis performed on multiparametric MRI have the potential to offer precise and comprehensive information on LF. This could facilitate early detection and timely intervention and reduce the reliance on invasive procedures such as liver biopsies. However, there is currently a lack of literature comparing and combining radiomics models based on multiparametric MRI sequences in rabbit models, potentially resulting in the omission of effective imaging techniques for predicting early-stage LF.
Therefore, the aim of this study was to compare the performance of individual radiomic models based on multiparametric MRI sequences in predicting early-stage LF and to identify the most effective radiomic models using a CCl4-induced rabbit model. Furthermore, a combined model was developed by integrating effective multiparametric radiomic features and validated to improve the accuracy of LF staging.
Methods
LF rabbit model establishment
This animal study received ethical approval (No. 2020-021) from our Committee. A total of 150 male New Zealand rabbits (aged 6 months, 2.5-3.0 kg) were included in this study. They were randomly scheduled to either induce LF (n = 120) or serve as normal rabbits (n = 30). LF rabbits received weekly subcutaneous injections in their backs with a 50% carbon tetrachloride (CCl4) oil solution.12 The injection volume varied based on the week: 0.1 ml/kg for weeks 1-3, 0.2 ml/kg for weeks 4-6, and 0.3 ml/kg for weeks 7-16. Conversely, the normal rabbits were administered subcutaneous injections of an equal volume of saline solution. Subsequently, both eligible normal and LF rabbits were enrolled in this study and randomly allocated into training and validation cohorts using a 7:3 cross-validation ratio.
Multiparametric MRI protocol
As the LF stages may advance with increasing doses of CCl4 and prolonged waiting times, a total of 30 rabbits per time point with LF and 7 to 8 normal rabbits per time point were randomly selected for MRI scans at specific intervals (ie, 4th, 8th, 12th, and 16th weekends) after the initial CCl4 administration to induce different stages of LF. This was conducted using a 3.0 T MRI scanner (Ingenia, Philips Healthcare, the Netherlands) equipped with a 20-channel head-neck coil. Prior to the MRI scan, the rabbits received general anaesthesia via intramuscular injection of 0.1 ml/kg xylazine. Throughout the MRI scan, the rabbits were placed in a supine position with their heads immobilized in a forward position. The multisequence MRI parameters, including MRE, DCE-MRI, HBP, IVIM, DKI and T2* sequences, are provided in the Supplementary material. To complete the MRE scan, a circular passive driver measuring 19 cm in diameter was positioned against the body wall over the liver and fastened using an elastic belt. DCE-MRI was continuously acquired, sampling 80 times at a rate of once every 8.66 s. Furthermore, a manual injection of 0.1 ml Gd-EOB-DTPA (Primovist, Bayer-Schering Pharma, Germany) was administered through the ear vein, followed by a 2 ml saline flush, and HBP was obtained 20 min later.
Histopathologic analysis
After the MRI examination, both the LF and normal rabbits were promptly euthanized at the corresponding time points, and liver tissue samples were harvested. Masson trichrome staining was performed on the liver samples, and a well-trained pathologist in hepatic pathology conducted the analysis. LF was analysed using the METAVIR scoring system,13 which categorizes LF into five stages: F0 (no fibrosis), F1 (portal fibrosis without septa), F2 (portal fibrosis with few septa), F3 (portal fibrosis with numerous septa but no cirrhosis), and F4 (cirrhosis). The stage with the highest LF among the liver tissues was considered the targeted stage. For further analysis, the LF stages were classified as follows: F0 to F2 as early-stage LF and F3 and F4 as advanced-stage LF.
Whole-liver ROI segmentation
Multiparametric images, including the MRE grayscale map, Ktrans map derived from DCE-MRI, HBP, D map derived from IVIM, and mean diffusion (MD) map derived from DKI and T2*, were imported into ITK-SNAP software (version 3.8.0). The detailed postprocessing protocol for multiparametric MRI images is shown in Supplementary Table S1. Radiologist 1 (MXF), who was blinded to the histopathological results, manually outlined the liver boundary on each consecutive transverse slice of the images, excluding artefacts, major vessels, bile ducts, and liver boundaries. This process led to the automatic generation of a 3D region of interest (ROI).
Radiomics feature extraction and dimension reduction
The Python package PyRadiomics (version 3.1.0) was utilized to extract various types of features, including shape (n = 17), first-order (n = 57), grey level cooccurrence matrix (n = 72), grey level dependence matrix (n = 42), grey level run length matrix features (n = 48), gray level size zone (n = 48), and neighbouring grey tone difference matrix (n = 15). As a result, 299 features were extracted from each MRI sequence for each individual rabbit.
To identify the optimal features, a four-step procedure was executed in a sequential manner. First, all the extracted features were subjected to normalization using the Z score method to minimize interference. Next, Spearman rank correlation was applied to eliminate features with a correlation coefficient greater than 0.9. Subsequently, a tenfold cross-validation technique was utilized in conjunction with the least absolute shrinkage and selection operator (LASSO) approach to identify the most accurate features with nonzero coefficients based on an optimal penalty parameter (λ). Finally, the maximum relevance minimum redundancy (mRMR) approach was employed to further decrease the dimensionality between the retained features.
Radiomics model development and verification
The selected robust features were weighted according to their respective coefficients to diagnose early-stage LF through the support vector machine (SVM) classifier, which contributed to establishing a radiomics model of MRE, DCE-MRI, HBP, IVIM, DKI, and T2*. Then, a linear logistic regression approach was used to calculate the quantitative Rad-scores of each model. Subsequently, univariable and multivariable logistic regression analyses were adopted to identify significant Rad-scores associated with early-stage LF. Afterward, the retained features were integrated to design the combined model. A comprehensive flowchart depicting the process from multiparametric scanning to the development of the radiomic model is presented in Figure 1.
Figure 1.
The detailed process from multiparmetric MRI scanning to model development. First, multiparmmetric MRI images were acquired, and the ROIs were segmented and then delineated. Then, whole-liver radiomics features were extracted. The Spearman correlation method, LASSO, and mRMR were used to reduce the number of dimensions and select features. Then, a support vector machine classifier was utilized to develop the radiomics model using pathological diagnosis as the reference standard. Finally, the ROC curve and DCA curve were used to assess the performance of various radiomics models for predicting early-stage LF.
Statistical analysis
Statistical analyses were performed using R software (version 4.0). The Shapiro-Wilk test was used to test the normality of the Rad-scores, which are presented as the mean ± standard deviation or median (interquartile range) following the normality assessment. The Mann-Whitney U test was used to detect differences between early-stage LF and advanced-stage LF. Univariable and multivariable logistic regression analyses were conducted to identify significant Rad-scores associated with early-stage LF. The diagnostic performance for predicting early-stage LF was evaluated through a receiver operating characteristic (ROC) curve, with assessments of sensitivity, specificity, accuracy, and area under the ROC curve (AUC). The Delong method was utilized to compare the AUC values from different models. Moreover, the Hosmer-Lemeshow test was used to obtain calibration curves, and decision curve analysis (DCA) was conducted to evaluate the goodness-of-fit and net benefits of all models in predicting early-stage LF.
Results
Animal model development and histopathological results
Among the 150 rabbits analysed, 12 LF rabbits did not survive due to difficulties tolerating CCl4 or anaesthesia. Additionally, 3 LF rabbits and 1 normal rabbit were excluded due to artefacts related to breathing and movement. In total, 29, 31, 31, 26, and 17 rabbits were ultimately diagnosed with stages F0, F1, F2, F3, and F4, respectively, based on histopathological examination. This included 91 rabbits with early-stage LF and 43 with advanced-stage LF. Within the training cohort, there were 20 rabbits with stage F0, 19 with stage F1, 21 with stage F2, 22 with stage F3, and 11 with stage F4 disease (early-stage LF, n = 60; advanced-stage LF, n = 33). The validation cohort consisted of 9, 12, 10, 4, and 6 rabbits diagnosed with stage F0, F1, F2, F3, and F4 disease, respectively (early-stage LF, n = 31; advanced-stage LF, n = 10).
Feature selection and Rad-score establishment
Redundant features with Spearman correlation coefficients exceeding 0.9 were subsequently removed, and 65, 54, 45, 47, 62, and 58 features were retained from the MRE, DCE, HBP, IVIM, DKI, and T2* models, respectively. After applying the LASSO and mRMR algorithms (Supplementary Figures S1-S14), a total of 10, 10, 13, 10, 12, and 6 highly robust features were selected to calculate the Rad-scores of the MRE, DCE-MRI, HBP, IVIM, DKI, and T2* models, respectively, according to the following formulas (Supplementary Table S2).
Comparison and logistic regression analysis of Rad-scores
As presented in Table 1, the early-stage LF group exhibited significantly greater Rad-scores than did the advanced-stage LF group among all multiparametric MRI images (P < .05). Additionally, the Rad-scores of the MRE, DCE, HBP, IVIM, DKI, and T2* models were found to be important factors significantly associated with early-stage LF. However, following multivariable logistic regression analysis, only the Rad-scores of MRE, DCE, and IVIM were found to be independently associated with early-stage LF. These three models were selected to develop three radiomics models, as depicted in Table 2.
Table 1.
Comparison of Radscores between early-stage LF and advanced-stage LF.
| Model name | All rabbits (n = 134) | Early-stage LF (n = 91) | Advanced-stage LF (n = 43) | P-value |
|---|---|---|---|---|
| MRE | 0.67 ± 0.28 | 0.81 ± 0.17 | 0.38 ± 0.21 | <.001 |
| DCE | 0.68 ± 0.17 | 0.76 ± 0.12 | 0.51 ± 0.14 | <.001 |
| HBP | 0.68 ± 0.21 | 0.77 ± 0.16 | 0.48 ± 0.17 | <.001 |
| IVIM | 0.68 ± 0.17 | 0.75 ± 0.14 | 0.52 ± 0.15 | <.001 |
| DKI | 0.69 ± 0.30 | 0.81 ± 0.20 | 0.42 ± 0.30 | <.001 |
| T2* | 0.67 ± 0.19 | 0.76 ± 0.15 | 0.49 ± 0.15 | <.001 |
MRE, magnetic resonance elastography; DCE-MRI, dynamic contrast-enhanced MRI; HBP, hepatobiliary phase; IVIM, intravoxel incoherent motion; DKI, diffusion kurtosis imaging; LF, liver fibrosis.
Table 2.
Univariate and multivariate logistic regression analysis for early-stage LF.
| Univariate analysis |
Multivariate analysis |
|||||
|---|---|---|---|---|---|---|
| Variables | OR | 95%CI | P-value | OR | 95%CI | P-value |
| MRE | 2139.97 | 213.25-2.15 × 105 | <.001 | 1.12 | 1.10-1.14 | .002 |
| DCE | 1.87 × 105 | 3742.25-9.37 × 105 | <.001 | 7654.02 | 34.50-1.70 × 106 | .001 |
| HBP | 3686.87 | 275.76-4.93 × 103 | <.001 | 0.00 | 0.00-25.21 | .197 |
| DKI | 143.70 | 28.57-722.81 | <.001 | 0.03 | 0.00-1.20 | .062 |
| IVIM | 2.47 × 104 | 775.53-788 × 105 | <.001 | 4.48 × 105 | 1.01-1.98 × 1011 | .048 |
| T2* | 1.1 × 104 | 574.47-2.12 × 105 | <.001 | 8.56 | 0.01-9.62 × 103 | .549 |
Individual model establishment
The MRE model had AUC values of 0.95 (0.91-0.99) in the training cohort and 0.86 (0.73-0.99) in the validation cohort, followed by the DCE model, which had AUC values of 0.93 (0.89-0.99) and 0.82 (0.64-0.99), respectively, and the IVIM model, which had AUC values of 0.91 (0.85-0.97) and 0.82 (0.68-0.97), respectively. The diagnostic ability of the radiomic models to predict early-stage LF is provided in Table 3 and illustrated in Figure 2A and B.
Table 3.
Diagnostic performance of the various model for predicting early-stage LF.
| Model name | Number of features | Cohort | Accuracy | AUC (95% CI) | Sensitivity | Specificity |
|---|---|---|---|---|---|---|
| MRE | 10 | Training | 0.90 | 0.95 (0.91-0.99) | 0.91 | 0.90 |
| Validation | 0.88 | 0.86 (0.73-0.99) | 0.89 | 0.85 | ||
| DCE | 10 | Training | 0.82 | 0.93 (0.89-0.99) | 0.73 | 1.00 |
| Validation | 0.83 | 0.82 (0.64-0.99) | 0.82 | 0.92 | ||
| IVIM | 10 | Training | 0.85 | 0.91 (0.85-0.97) | 0.83 | 0.90 |
| Validation | 0.83 | 0.82 (0.68-0.97) | 0.82 | 0.85 | ||
| Combined | 19 | Training | 0.85 | 0.96 (0.92-0.99) | 0.79 | 0.97 |
| Validation | 0.88 | 0.88 (0.77-0.99) | 0.93 | 0.77 |
Figure 2.
The ROC, calibration and DCA curves of all the radiomic models for predicting early-stage LF. The combined model presented the highest AUC values of 0.96 and 0.88 in the training (A) and validation (B) cohorts, respectively. Moreover, the combined model curve was well fitted with a perfectly calibrated curve (C and D) and was greater than either all or none across the most reasonable threshold probabilities (E and F).
Combined model development and application assessment
Through the combination of effective features extracted from the MRE, DCE, and IVIM models, the combined model achieved an AUC of 0.96 (0.92-0.99) in the training cohort and 0.88 (0.77-0.99) in the validation cohort. According to the Delong comparison results (Supplementary Figures S15 and S16), the combined model exhibited a significantly greater AUC than did the IVIM model (P < .05). The sample prediction histograms of the MRE, DCE and combined models for predicting early-stage LF are shown in Figure 3.
Figure 3.
Sample prediction histograms of the MRE, DCE, and combined radiomics models for predicting early-stage LF in the training and validation cohorts.
The Hosmer-Lemeshow results (Figure 2C and D) indicated that the curves for all four models exhibited a good fit to the perfectly calibrated curve, with the combined models demonstrating excellent agreement in both cohorts. Consequently, there was a strong correspondence between the predictive model and early-stage LF diagnosis. Additionally, according to the DCA curve, all model curves consistently exhibited nonlinear trends across a wide range of threshold probabilities, particularly for the combined model, indicating increased net benefit for predicting early-stage LF (Figure 2E and F).
Discussion
Animal research is essential for advancing clinical applications and significantly contributes to the widespread adoption of MRI technology, specifically in the detection of LF. Our study aimed to assess six MRI sequences to determine the most accurate and standardized quantitative imaging techniques for diagnosing LF progression. The analysis revealed that Rad-scores from the MRE, DCE, and IVIM models were independent predictors associated with early-stage LF, with the MRE model exhibiting particularly superior performance in LF diagnosis. Furthermore, the combined model based on multiparametric MRI exhibited remarkable performance, suggesting its potential as an effective tool for individualized imaging to predict early-stage LF.
Functional MRI sequences provide valuable information about the metabolic and physiological changes that occur during LF progression.14 MRE is commonly considered the most effective imaging procedure for LF diagnosis and staging and can even serve as the gold standard due to its similar performance to that of liver biopsy.15,16 One of the main reasons for its effectiveness is that MRE is highly sensitive in detecting liver stiffness, which increases as LF severity progresses. In this study, the Rad-scores derived from MRE were suggested to be independent predictors of LF progression and exhibited superiority, with an AUC value of 0.95 in diagnosing early-stage LF compared to other individual radiomic models. Sim et al17 reported a concordant AUC value of 0.96 for detecting significant hepatic fibrosis (stage F2-F4) in patients using an MRE radiomics model. This finding reinforces and expands the application of MRE in LF staging, despite the use of different LF classifications (stage ≥ F2) in the human setting.
Both Zou et al12 and Liu et al18 reported that the Ktrans parameter derived from DCE-MRI serves as a valuable imaging biomarker for evaluating microvascular permeability and plasma flow perfusion and therefore exhibited greater accuracy in diagnosing and staging LF. Based on these findings, we selected the Ktrans map to extract features and develop a DCE-MRI model. The resulting model achieved an AUC of 0.93 for predicting early-stage LF. Nevertheless, it is worth noting that Zou et al12 found relatively low diagnostic performance of the quantitative parameters of Ktrans measured on DCE-MRI images, with AUC values ranging from 0.54 to 0.89 for staging LF. This finding reinforces the value of the radiomic model in improving diagnostic performance for predicting LF, as it allows the extraction and combination of numerous invisible features to design a more accurate radiomic model.19
Intravoxel incoherent motion is a biexponential model that utilizes diffusion weighted imaging (DWI) to separately assess true molecular diffusion and perfusion status without the need for contrast agent injection, making it useful for liver function assessment, LF staging, and tumour characterization.20 A meta-analysis conducted by Ye et al21 involving 1388 patients demonstrated a pooled AUC of 0.89 (0.87-0.91) for detecting advanced-stage LF. Moreover, studies by Tosun et al22 and Chung23 et al showed that the D value derived from IVIM exhibited superior diagnostic performance in discriminating LF stages compared to other IVIM-derived or DWI-derived parameters. In the present study, we aimed to develop an IVIM-D map-based radiomic model that showed a slightly greater AUC value of 0.91 for predicting early-stage LF, possibly because the radiomic model can capture subtle features that may be indicative of early-stage LF.
Considering the pharmacokinetic properties of Gd-EOB-DTPA, HBP imaging has potential for evaluating liver function because the signal intensity of HBP decreases as the stage of LF worsens.24 However, the Rad-scores of the HBP radiomics model were not included in the multivariable logistic regression analyses, contradicting a previous study by Park et al,25 which reported an AUC of 0.88 for detecting advanced-stage LF in patients. Notably, Park et al focused on drawing ROIs only in the right hepatic lobe, and the use of different radiomic features might lead to disparate outcomes. The DKI26 and T2*27 sequences assess the kurtosis of the water diffusion probability distribution function and iron deposition, respectively, during the progression of LF. Nevertheless, the Rad-scores of the DKI and T2* models did not emerge as independent factors associated with early-stage LF. This is mainly attributed to the fact that the underlying pathologies of LF, inflammation activity28 and iron deposition29 play significant roles. Iron deposition is theoretically expected to decrease T2* times due to its paramagnetic nature, while the impact of inflammation might interfere with the water molecular motion detected by DKI, consequently diminishing the diagnostic efficacy of these two radiomic models in distinguishing between stages of LF.
The integration of multiparametric radiomic features to develop combined models has been the subject of several studies aiming to achieve greater diagnostic accuracy. For instance, Wei et al30 established a combined model based on unenhanced T1WI and T2WI images, which demonstrated higher AUC values for the discrimination of significant LF stages and grading inflammatory activity. In our study, univariable and multivariable logistic regression analyses were sequentially conducted to identify significant Rad-scores associated with early-stage LF. Afterwards, these effective MRI sequences were combined to develop a multiparametric MRI model with improved performance. To our knowledge, our study is the first to produce a combined model based on radiomic features extracted from MRE, DCE, and IVIM sequences. The combined model exhibited superior performance compared to the single radiomics model, with a greater AUC of 0.96 (0.92-0.99) for predicting early-stage LF. This finding highlights the potential value of the complementarity and collaboration of multiparametric features in improving the ability to discriminate LF stages. Furthermore, calibration curves and DCA curves indicated that all predictive models, particularly the combined model, demonstrated better goodness-of-fit and increased net benefit, further highlighting its stable and valuable performance in predicting early-stage LF.
Previous studies have increasingly utilized a wide array of advanced MRI sequences, such as MRE, HBP, DCE-MRI, IVIM, DKI, and T2*, for staging and assessing LF.27–30 These investigations have revealed alternative noninvasive methods for LF staging. Nevertheless, a direct comparison to determine the most effective imaging technique has not been conducted. Additionally, no whole-liver radiomic studies based on multiparametric MRI sequences have been performed or compared. In this study, a whole-liver radiomic model was established and compared to select the most effective MRI sequence. This model was further integrated to establish a combined model, which is beneficial for guiding clinical decisions.
Our study has several limitations that should be acknowledged. First, due to limited access to pathological factors from histopathology, our study focused solely on examining the correlation between radiomics models and LF stage. Therefore, future studies that account for adjustments for inflammation and iron deposition in the liver are necessary to improve the reliability of our study results. Second, radiomic features were only extracted from the Ktrans map, D map, and MD map, which potentially overlook valuable features from other MRI sequences. This performance is not only for extracting the most effective radiomic model but also for avoiding overfitting caused by many radiomic features. Moreover, it is worth noting that a rabbit model induced by CCl4 may not accurately represent pathological alterations in the human liver. Hence, additional investigations in a clinical environment are imperative to validate the outcomes of this research.
Conclusion
In summary, the MRE, DCE, and IVIM radiomics models are valuable for noninvasively assessing LF, and the MRE model outperformed the DCE and IVIM models. Furthermore, the combination of these three models has the potential to further improve the predictive accuracy and serve as an excellent imaging procedure for the early-stage prediction of LF.
Supplementary Material
Contributor Information
Xiao-Fei Mai, Department of Radiology, Sixth Affiliated Hospital of Shenzhen University, Shenzhen 518052, China.
Hao Zhang, Department of Radiology, Sixth Affiliated Hospital of Shenzhen University, Shenzhen 518052, China.
Yang Wang, Department of Radiology, Sixth Affiliated Hospital of Shenzhen University, Shenzhen 518052, China.
Wen-Xin Zhong, Department of Radiology, Sixth Affiliated Hospital of Shenzhen University, Shenzhen 518052, China.
Li-Qiu Zou, Department of Radiology, Sixth Affiliated Hospital of Shenzhen University, Shenzhen 518052, China.
Supplementary material
Supplementary material is available at BJR online.
Funding
This work was supported by the National Natural Science Foundation of China (No. 81771805).
Conflicts of interest
The authors declare that they have no conflict of interest.
Ethical approval
This study was approved by Ethics Committee of Sixth Affiliated Hospital of Shenzhen University.
References
- 1. Kisseleva T, Brenner D.. Molecular and cellular mechanisms of liver fibrosis and its regression. Nat Rev Gastroenterol Hepatol. 2021;18(3):151-166. [DOI] [PubMed] [Google Scholar]
- 2. Aydin MM, Akcali KC.. Liver fibrosis. Turk J Gastroenterol. 2018;29(1):14-21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Schuppan D, Ashfaq-Khan M, Yang AT, Kim YO.. Liver fibrosis: direct antifibrotic agents and targeted therapies. Matrix Biol. 2018;68-69:435-451. [DOI] [PubMed] [Google Scholar]
- 4. Jung YK, Yim HJ.. Reversal of liver cirrhosis: current evidence and expectations. Korean J Intern Med. 2017;32(2):213-228. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Karvellas CJ, Francoz C, Weiss E.. Liver transplantation in acute-on-chronic liver failure. Transplantation. 2021;105(7):1471-1481. [DOI] [PubMed] [Google Scholar]
- 6. Khalifa A, Rockey DC.. The utility of liver biopsy in 2020. Curr Opin Gastroenterol. 2020;36(3):184-191. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Liu HF, Lu Y, Wang Q, et al. Machine learning-based CEMRI radiomics integrating LI RADS features achieves optimal evaluation of hepatocellular carcinoma differentiation. J Hepatocell Carcinoma. 2023;13:2103-2115. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Liu HF, Zhang YZZ, Wang Q, et al. A nomogram model integrating LI-RADS features and radiomics based on contrast-enhanced magnetic resonance imaging for predicting microvascular invasion in hepatocellular carcinoma falling the Milan criteria. Transl Oncol. 2023;27:101597. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Venkatesh SK, Torbenson MS.. Liver fibrosis quantification. Abdom Radiol (NY). 2022;47(3):1032-1052. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Im WH, Song JS, Jang W.. Noninvasive staging of liver fibrosis: review of current quantitative CT and MRI-based techniques. Abdom Radiol (NY). 2022;47(9):3068-3067. [DOI] [PubMed] [Google Scholar]
- 11. Welle CL, Olson MC, Reeder SB, et al. Magnetic resonance imaging of liver fibrosis, fat, and iron. Radiol Clin North Am. 2022;60(5):705-716. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Zou L, Jiang J, Zhang H, et al. Comparing and combining MRE, T1ρ, SWI, IVIM, and DCE-MRI for the staging of liver fibrosis in rabbits: assessment of a predictive model based on multiparametric MRI. Magn Reson Med. 2022;87(5):2424-2435. 10.1002/mrm.29126 [DOI] [PubMed] [Google Scholar]
- 13. Goodman ZD. Grading and staging systems for inflammation and fibrosis in chronic liver diseases. J Hepatol. 2007;47(4):598-607. [DOI] [PubMed] [Google Scholar]
- 14. Petitclerc L, Sebastiani G, Gilbert G, et al. Liver fibrosis: review of current imaging and MRI quantification techniques. J Magn Reson Imaging. 2017;45(5):1276-1295. [DOI] [PubMed] [Google Scholar]
- 15. Kim YS, Jang YN, Song JS.. Comparison of gradient-recalled echo and spin-echo echo-planar imaging MR elastography in staging liver fibrosis: a meta-analysis. Eur Radiol. 2018;28(4):1709-1718. [DOI] [PubMed] [Google Scholar]
- 16. Ozturk A, Olson MC, Samir AE, et al. Liver fibrosis assessment: MR and US elastography. Abdom Radiol (NY). 2022;47(9):3037-3050. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Sim KC, Kim MJ, Cho Y, et al. Diagnostic feasibility of magnetic resonance elastography radiomics analysis for the assessment of hepatic fibrosis in patients with nonalcoholic fatty liver disease. J Comput Assist Tomogr. 2022;46(4):505-513. [DOI] [PubMed] [Google Scholar]
- 18. Liu HF, Wang Q, Du YN, et al. Dynamic contrast-enhanced MRI with Gd-EOB-DTPA for the quantitative assessment of early-stage liver fibrosis induced by carbon tetrachloride in rabbits. Magn Reson Imaging. 2020;70:57-63. [DOI] [PubMed] [Google Scholar]
- 19. Rogers W, Thulasi Seetha S, Refaee TAG, et al. Radiomics: from qualitative to quantitative imaging. Br J Radiol. 2020;93(1108):20190948. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Li YT, Cercueil JP, Yuan J, et al. Liver intravoxel incoherent motion (IVIM) magnetic resonance imaging: a comprehensive review of published data on normal values and applications for fibrosis and tumor evaluation. Quant Imaging Med Surg. 2017;7(1):59-78. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Ye Z, Wei Y, Chen J, et al. Value of intravoxel incoherent motion in detecting and staging liver fibrosis: A meta-analysis. World J Gastroenterol. 2020;26(23):3304-3317. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Tosun M, Onal T, Uslu H, et al. Intravoxel incoherent motion imaging for diagnosing and staging the liver fibrosis and inflammation. Abdom Radiol (NY). 2020;45(1):15-23. [DOI] [PubMed] [Google Scholar]
- 23. Chung SR, Lee SS, Kim N, et al. Intravoxel incoherent motion MRI for liver fibrosis assessment: a pilot study. Acta Radiol. 2015;56(12):1428-1436. [DOI] [PubMed] [Google Scholar]
- 24. Wang C, Yuan XD, Wu N, Sun WR, Tian Y.. Optimization of hepatobiliary phase imaging in gadoxetic acid-enhanced magnetic resonance imaging: a narrative review. Quant Imaging Med Surg. 2023;13(3):1972-1982. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Park HJ, Lee SS, Park B, et al. Radiomics analysis of gadoxetic acid–enhanced MRI for staging liver fibrosis. Radiology. 2019;290(2):380-387. [DOI] [PubMed] [Google Scholar]
- 26. Yang L, Rao S, Wang W, et al. Staging liver fibrosis with DWI: is there an added value for diffusion kurtosis imaging? Eur Radiol. 2018;28(7):3041-3049. [DOI] [PubMed] [Google Scholar]
- 27. Guimaraes AR, Siqueira L, Uppal R, et al. T2 relaxation time is related to liver fibrosis severity. Quant Imaging Med Surg. 2016;6(2):103-114. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Sheng RF, Wang HQ, Yang L, et al. Diffusion kurtosis imaging and diffusion-weighted imaging in assessment of liver fibrosis stage and necroinflammatory activity. Abdom Radiol (NY). 2017;42(4):1176-1182. [DOI] [PubMed] [Google Scholar]
- 29. Zou LQ, Liu HF, Du YN, et al. Effect of iron deposition on native T1 mapping and blood oxygen level dependent for the assessment of liver fibrosis in rabbits with carbon tetrachloride intoxication. Acad Radiol. 2023;30(5):873-880. [DOI] [PubMed] [Google Scholar]
- 30. Wei H, Shao Z, Fu F, et al. Value of multimodal MRI radiomics and machine learning in predicting staging liver fibrosis and grading inflammatory activity. Br J Radiol. 2023;96(1141):20220512. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.