Skip to main content

The application of texture quantification in hepatocellular carcinoma using CT and MRI: a review of perspectives and challenges

Abstract

Recently, radiomic texture quantification of tumors has received much attention from radiologists, scientists, and stakeholders because several results have shown the feasibility of using the technique to diagnose and manage oncological conditions. In patients with hepatocellular carcinoma, radiomics has been applied in all stages of tumor evaluation, including diagnosis and characterization of the genotypic behavior of the tumor, monitoring of treatment responses and prediction of various clinical endpoints. It is also useful in selecting suitable candidates for specific treatment strategies. However, the clinical validation of hepatocellular carcinoma radiomics is limited by challenges in imaging protocol and data acquisition parameters, challenges in segmentation techniques, dimensionality reduction, and modeling methods. Identification of the best segmentation and optimal modeling methods, as well as texture features most stable to imaging protocol variability would go a long way in harmonizing HCC radiomics for personalized patient care. This article reviews the process of HCC radiomics, its clinical applications, associated challenges, and current optimization strategies.

Introduction

Hepatocellular carcinoma (HCC) is characterized by an increasing incidence, higher mortality, and morbidity burden. Currently, it is the second most common cause of cancer-related death worldwide, with about 50 % of all cases occurring in China alone [1]. The high mortality and morbidity burden mostly results from the late presentation of the disease in the majority of patients. In most cases, the diagnosis and follow-up of hepatocellular carcinoma, especially in the setting of cirrhosis can be effectively achieved using computed tomography (CT) and magnetic resonance (MR) imaging without the need for histological confirmation [2]. In recent years, there is an increasing need for better characterization of tumor heterogeneity and prediction of survival outcomes to permit individualized patient care. In this regard, the conventional qualitative CT and MR imaging modalities have fallen short, and consequently, the growing demand for more objective quantification of texture features.

Quantitative texture analysis of medical images – considered as a virtual biopsy technique – explores the microscopic details of a tumor by taking advantage of the spatial distribution and variation in the gray-level intensities of the pixels/voxels that make up the tumor images [3]. It has been expanded into the field of radiomics, which is an evolving research area that entails extracting large texture data to generate predictive models for prognostication, better tumor characterization, and assessing treatment responses. Although the Barcelona Clinic Liver Cancer (BCLC) is still the most popular treatment guideline, HCC radiomics analysis has proven useful in the guidance of treatment strategies and the prediction of therapy response. Despite its convenience and advantages, the full clinical application of radiomics analysis has been hampered by the absence of a standard execution guideline, delaying the clinical validation of the aforementioned applications. Promising studies on HCC applied radiomics differently in terms of the imaging protocol, segmentation methods, and model construction. However, for patients to fully benefit from a clinically validated radiomics, certain challenges about HCC texture quantification need to be addressed in the multi-step process of radiomics. This article explores the prospects and challenges in data acquisition, segmentation and modeling methods by reviewing recent developments in the clinical applications of HCC radiomics, which is vital for guiding its clinical applications to ensure a personalized medicine with optimized treatment strategies for better patient prognosis.

Data acquisition and challenges

Most texture quantification of HCC is carried out using CT or MRI, as indicated in previous literature [4,5,6]. The universal problem associated with texture quantification in all available CT and MRI imaging modalities is in choosing the best imaging protocol, tumor segmentation methods, stable texture parameters, and the radiomic software tools used for analysis. In clinical work, different authors use different imaging parameters, depending on the equipment available in their institutions, which can account for the heterogeneity in tumor quantification [7, 8]. However, not all imaging parameters significantly affect the robustness of texture features; some vital parameters such as variations in scanner model, contrast injection rates, pixel resolutions, signal-to-noise ratio (SNR), and reconstruction algorithms have an obvious influence on the quantification of HCC texture feature. Therefore, it is important to develop a multi-parameter model that automatically corrects the variation in the various key data-acquisition parameters, or establish a comprehensive algorithm with a controlled imaging protocol that can provide stable texture features and enable an objective comparison between HCC radiomics studies, thereby promoting its clinical validation.

CT-based radiomics

In general, CT-based radiomics studies have shown that the variation of image acquisition parameters such as slice thickness, reconstruction algorithms, image resolution, contrast medium, and scanner type has the most significant influence on texture quantification [9,10,11]. In particular, reconstruction algorithms, pixel resolution, changes in contrast injection rates, and scanner models have been specifically implicated in HCC radiomics [9, 10, 12]. Regarding slice thickness, thinner image slices (1.25 and 2.5 mm) yield more quantitative texture information than thick slices (5 mm) [11]. Furthermore, the reconstruction algorithms: adaptive statistical iterative reconstruction (ASIR), model-based iterative reconstruction (MBIR) and the filtered back-projection (FBP) have each been shown to affect the quantification of certain texture features of liver lesions (p < 0.002) [10]. And in patients who had a retest contrast-enhanced CT scan 2 weeks apart, Perrin et al. showed that the variations in pixel resolutions and contrast injection rates could affect the number of reproducible texture features (gray-level co-occurrence matrix, GLCM; gray-level run-length matrix, GLRLM; intensity histogram and local binary patterns) between test and retest scans (concordance correlation coefficient of > 0.9) [9]. Also, each scanner model comes with a unique built-in pitch setting that can also influence the type of radiomics features extracted by the scanner. Finally, all other things being equal, the change in tube current and voltage [13], and the radiation dose [10] have no significant impact on HCC texture feature quantification.

MRI-based radiomics

In contrast to CT, determining the effect of data acquisition parameters on MR texture features can be more complicated because the imaging parameters, as well as contrast agents, bear no linear relationship with the signal intensity [14]. Generally, changes in echo time (TE), repetition time (TR), sampling bandwidth (SBW), spatial resolution, signal-to-noise ratio (SNR), field strength, scanner model, reconstruction algorithm, and parallel imaging acceleration factor, have been implicated in MRI-based radiomics [15,16,17]. Mayerhoefer et al. demonstrated that at a higher resolution, TE, TR, and SBW had a negligible effect on GLCM derived features in a phantom study using polystyrene spheres and an agar gel solution (PSAG) [15]. Generally, increasing spatial resolution and SNR improves the exploration of tumor heterogeneity [18]. In contrast, variation in the slice thickness of MR images doesn’t significantly affect the robustness of texture features [19, 20].

CT VS MRI-based radiomics

The superior diagnostic performance of MRI compared to CT in HCC has been documented [21] but their performance in radiomics analysis, however, has not been compared. As MRI has relatively higher sensitivity with a better spatial resolution and soft-tissue characterization, it might offer more robust texture features for tumor heterogeneity assessment than the CT [22]. A recent study compared the repeatability of CT and MR (using volumetric interpolation breath-hold examination (VIBE) and true fast MRI with steady state precession (TRUFISP) texture features of non-small cell lung cancer and showed 12 significant models that accurately predict overall survival but not tumor response. CT and MRI had a fairly similar predictive accuracy; 54.4% of CT texture features, 64.4% of TRUFISP and 52.6% of VIBE texture features were reproducible with a concordance correlation coefficient of ≥0.9 [23]. However as mentioned previously, simulation of the ground-truth textural composition of tissues of MR images can be more difficult, since the image signal intensities of tissues are strongly influenced by the MR acquisition parameters; moreover, images are more prone to artifacts that affect the quantitative analysis of texture features (especially the Gibbs ringing) compared to CT [14]. Thus, MRI-based radiomics signature may likely be more predictive of tumor heterogeneity but might be more susceptible to variations in imaging parameters compared to CT-based radiomics.

For both CT and MRI-based radiomics, good reproducibility of texture quantification is vital to the texture quantification. The reproducibility of the texture features is in turn affected by the imaging protocol and parameters.

Therefore, to improve the reproducibility of texture features, it is essential to understand how the alterations in the imaging protocol affect different texture features to allow the selection of suitable features under different imaging parameters (Table 1 below). This is because certain tumor texture features are less susceptible to even the most influencing parameter alterations. For example, some phantom studies [15,16,17] have demonstrated that GLCM-based features have greater reproducibility as they are more stable to variations in imaging parameters. Further studies using HCC phantoms are needed to determine if similar findings would be obtainable for hepatic malignancies.

Table 1 Studies that evaluated the impact of variation in imaging parameters on HCC texture quantification

The process of HCC radiomics

Texture quantification has been outlined in length in various studies [24,25,26]. Various software tools and statistical algorithms could be used on extracted image features to interpret tumor characteristics objectively and to determine the relationship between the tumor surface and the surrounding parenchyma [27]. A vast amount of texture features (currently around 50–5000) [28] from tumors can be converted into a mineable data from which radiomics and delta-radiomics signatures can be developed and then utilized for individualized patient care [29]. The process of HCC radiomics analysis begins with segmentation followed by feature selection, construction of radiomics signature, model building, and validation [4, 6] (Fig. 1).

Fig. 1
figure 1

schematic diagram summarizing the steps in HCC radiomics

Segmentation

The segmentation of the whole or part of a tumor using a radiomic software on a delineated region or volume of interest (ROI/VOI) for extraction is the first step in analyzing the texture of an image. The extracted features are described as semantic and quantitative. Segmentation can be manual, semiautomatic, or automatic. However, the semiautomatic segmentation method is currently more acceptable because it is less associated with bias from intra- or inter-observer variability compared with the manual delineation [8, 30] and less complex than the automatic method [31]. A study comparing the inter-observer variability of manual and semiautomatic segmentation showed that the semiautomated algorithm showed less variability compared with manual delineations (an interclass coefficient of 0.856 vs 0.776) [30]. The automated algorithms are limited by computational complexity, heterogeneity in HCC sizes and shape, and the need for a large amount of data set [30].

Current semiautomated segmentation algorithms proposed in the literature are broadly categorized into two: image intensity and tumor border-based methods [14]. The contour-based algorithms utilize the tumor contour detail while the intensity-based algorithms employ the tumor-parenchyma intensity gradient difference to execute segmentation. Algorithms that use tumor boundary information or intensity differences to segment tumors include region-growing (like the GrowCut), GraphCut, watersheds, livewires and active contours. GrowCut and GraphCut have become popular methods for semiautomated segmentation. Table 2 below summarizes the semiautomated segmentation algorithms.

Table 2 common semiautomated segmentation algorithms used in HCC

Semantic features

Semantic features are the “apparent” tumor radiological features observed in daily practice. They include tumor location, size, and shape; enhancement characteristics, effusions, etc. [36]. Segal et al. used 28 semantic CT texture features to decode 78% of the genes expressed in liver cancers [37]. Similarly, two other studies used semantic features including liver-tumor interface difference and hypointense halo to develop a radiomic signature, which to some extent predicted microvascular invasion (MVI) of HCC [38, 39]. Although less affected by variations in imaging protocol, the semantic features suffer more subjectivity and variability, thus limiting their robustness [40]. However, they are still very useful, especially when combined with the quantitative features [41, 42].

Quantitative features

Quantitative features are those tumor characteristics that cannot be seen by mere observation of images; they require deciphering using various approaches including statistical-, model-, transform-based and structural methods [43, 44]. By and large, the statistical-based method is widely used in clinical practice for quantitative texture analysis. Quantitative features in the statistical model are grouped into first-order, second-order and higher-order statistics (Table 3), which have been used to represent the texture feature, and the second and higher-order statistics are of high importance for the evaluation of tumor characteristics in many HCC studies.

Table 3 The summary of the statistical model used in texture quantification

Challenges in tumor segmentation

A big challenge in tumor segmentation is in choosing the part of the tumor to be used in mining the radiomic data. Different studies take a different approach to perform tumor segmentation; some authors use part of the tumor while others use the whole tumor. Furthermore, other authors included the tumor and peritumoral region, and recently, Blanc-Durand et al. segmented the entire liver to build a whole-liver radiomics [47]. Francesca et al. showed a texture analysis of the entire tumor to yield more feature information, because texture characteristics of a particular ROI in the tumor, just like a biopsy sample, may not be the true representation of the entire tumor heterogeneity [48]. While some authors advocated exclusive segmentation of the tumor [5], more recently, others extended into the adjacent peritumoral tissue to generate a combined intra and peritumoral ROI/VOI. The latter approach has been shown to provide more detail on tumor heterogeneity, especially about MVI status [29, 49, 50]. A recent study demonstrated that variations in the methods of tumor segmentation affected the quantification and robustness of tumor texture features and emphasized the need for adopting a segmentation method from which the most stable radiomics features could be extracted [8].

Radiomic model

After segmentation of the ROI/VOI, appropriate texture features are selected to build, train, and validate a predictive radiomic model.

Feature selection

Because the number of texture features obtained from a tumor can be quite large, the purpose of the feature selection step is to select only clinically relevant texture features that would be incorporated into a radiomic model. Extracted features are often narrowed to not more than the sample size [51] to avoid the curse of dimensionality, which occurs when the model is over-fitted with so many features: including redundant and irrelevant features. This affects the model’s performance on a new dataset due to high variance, leading to erroneous predictions [52]. There are various feature selection methods [53] but the most commonly employed approach is the filtration technique that is used to remove noise from selected features by highlighting texture features of specified size on the spatial scaling factor (ranging between 2 and 6 mm) [54]. Relevant and stable filtered features are computed by statistical methods including intraclass correlation coefficient (ICC), Students t-test, Mann-Whitney U test, etc. [26].

Model building and training

Since there are no established standard guidelines, researchers use different methods to build predictive models; however, regardless of the modeling method applied, the technique should be fully documented in a pellucid and easily reproducible manner [24]. Modeling can generally be supervised, semi-supervised or unsupervised. The commonly adopted method is the supervised approach in which a model is trained to characterize tumor heterogeneity or predict an outcome by feeding it with specific ground truth clinical labels and then tested for its performance [26]. If all the different models had the same predictive performance, comparison between studies would have been easier; however, because different modeling methods have different predictive accuracies, there is a need to identify the model with the best performance to achieve the desired clinical endpoint. On this regard, Ni et al. demonstrated that a combination of the LASSO+GBDT models (least absolute shrinkage and selection operator plus gradient boosting decision tree) had a higher diagnostic accuracy (highest area under the curve) for MVI of HCC compared to the back-propagation neural network (BPNet), K nearest neighbors (KNN), support vector machine (SVM), random forest (RF), decision tree (DT), and Bayesian models [55]. Moreover, recent studies on HCC quantitative analysis have shown that a combined radiomic nomogram composed of clinical, laboratory, semantic, and radiomic signatures, has better predictive power than a single radiomic model (Fig. 2). This is because variables such as the alpha-fetoprotein levels (AFP level), Child-Pugh score, and HBV status influence tumor heterogeneity [42]. For example, Kim and colleagues [42] improved the performance of their radiomic model in predicting survival form a hazard ratio of 7.42 to 19.88 (p < 0.0001) by incorporating alpha-fetoprotein (AFP) levels, liver function status (Child-Pugh score) and tumor size. Similarly, high AFP (> 400 ng/ml) and AST (> 40 U/L) levels are two clinical parameters that independently predicted MVI of HCC with an accuracy of 72.4 and 65.5%, respectively. When combined with radiomic signatures, the accuracy of the combined model improved to 82.8% (AUC of 0.889) [29]. Another study showed that both clinical (consisting of patients age, AFP, HBsAg, and tumor size) and radiomic models performed poorly in predicting MVI (AUC of 0.734 and 0.783, respectively). But when combined together, the performance of the clinical-radiomic significantly improved (to AUC of 0.835) [56].

Fig. 2
figure 2

Illustration of a radiomic nomogram using clinical, laboratory and radiomics signature, AFP = alpha-fetoprotein, HBV = hepatitis B virus [56]

Validation

After generating a suitable model, the final step is to validate its capability in accurately predicting the desired clinical outcome for which it is being built. Models can be validated internally by either using split or cross-validation methods, or more preferably externally using independent patient cohorts not included in the model construction. Several statistical methods are used to validate a model’s performance including the concordance index (C-index) and time-dependent receiver operating characteristic curve (ROC), which are generally used for models built to predict survival outcomes. For models built to predict a particular event such as MVI; the area under the curve (AUC), sensitivity, specificity, and calibration are utilized for validation [26] (summarized in Table 4 below). The best model is the combined model (clinical-radiomic model) with the maximum AUC values in most studies and should be recommended in texture quantification analysis due to its highest performance.

Table 4 summary of studies showing the predictive performance of radiomics signature, clinical-radiological and the combined models

Challenges

The performance of models varies because each modeling technique has its unique limitations. Thus, the main challenge associated with modeling is the dare need for selecting an appropriate method for a particular event to be predicted [55]. Notably, the supervised learning models often require large amounts of training variables, which sometimes may not be enough to achieve optimal training to permit selection of the most relevant features. For this reason, a semi-supervised modeling method can be used to respond to insufficient data labels [26].

Clinical application

Texture analysis of CT and MR images using tumor alone, a combination of the tumor and peritumoral region, or the whole liver (including all lesions) to generate a single radiomics signature or a combined clinical-pathologic-radiomic model has been applied to characterize and grade HCCs, assess MVI, monitor treatment response, and predict clinical endpoints.

Tumor characterization

Texture quantification of CT and MR images has proven quite useful in the characterization of liver lesions and particularly in atypical HCC in the non-cirrhotic liver where the diagnosis can be challenging even with the liver-specific contrast-enhanced MRI [61]. Stocker et al. used GLRLM, GLCM and gray-level histogram derived features to differentiate between HCCs and benign lesions in non-cirrhotic livers [12]. Texture features from spectral attenuated inversion recovery (SPAIR) T2W images have also been used to differentiate HCC from hepatic hemangioma and liver metastasis [62]. Also, a combined clinical and radiomic model has been used to predict MVI in HCC [58]. Moreover, some authors have used texture analysis to assign specific treatment options to patients based on their tumor textural characteristics [63, 64].

Response assessment

Texture quantification has been successfully used in monitoring response after both surgical and locoregional treatment of HCC. By quantifying change in tumor heterogeneity, radiomics analysis has the potential of assessing HCC’s response to treatments; however, only a few studies explored this potential [46]. Kloth et al. compared CT texture analysis with modified response evaluation criteria in solid tumors (mRECIST) and perfusion CT in assessing tumor response after drug eluting-bead transarterial chemoembolization (DEB-TACE) and showed that texture quantification could augment perfusion CT and mRECIST in monitoring treatment response [65]. Additionally, Yu JY et al. compared the texture features of HCC before and after TACE combined with high intensity focused ultrasound (HIFU) therapy and concluded that skewness (AUC 0.76, p < 0.05) and entropy (AUC 0.736, p < 0.05) from arterial phase images a week after treatment were strong predictors of early response [66].

Prediction of survival

Texture analysis of HCC has been used to predict the survival outcomes of patients after various treatment strategies [47, 63, 67, 68]. A study showed that texture characteristics could be used to select patients for combined TACE plus sorafenib, as well as to predict survival; Gabor-1-90 (filter 0) and wavelet-3-D (filter 1.0) from portal phase CT images were predictive of time to progression (TTP) and overall survival (OS), respectively. They proposed that patients having lower Gabor-1-90 (filter 0) and wavelet-3-D (filter 1.0) would benefit from TACE plus sorafenib [63]. Similarly, preoperative texture features from MR images have been used to predict early recurrence of HCC after hepatectomy. The study demonstrated that entropy and uniformity from the arterial phase images were independent predictors of early recurrence in patients with tumors ≤3 cm (p = 0.031 and p = 0.014, respectively) while entropy and skewness from same phase images were independent predictors of early recurrence in tumors > 3 cm [22]. And to predict response after chemotherapy with sorafenib, entropy from CT portal venous phase images was significantly (p < 0.05) associated with overall survival in both training and validation cohorts [68].

Some of the vital clinical applications of HCC radiomics are summarized in Table 5 below, and the most robust texture features are obtained from the second and higher-order statistics, which are GLCM, GLRLM, Gabor, and wavelet transform.

Table 5 Summary of the studies on radiomics analysis of HCC

Relationship between the radiomics features and Histopathologic correlates

In principle, tumor grading and immunotyping are determined by histopathological examination. Linking robust radiomics features with specific tumor histological markers will improve clinical decision-making without resorting to invasive procedures. But translating this correlation to tumor pathogenesis and outcome is difficult due to a large number of texture features and heterogeneity in HCC radiomics studies.

Zhou et al. linked MRI-based mean intensity and GLRLM-nonuniformity to histologic grades of HCC. Low-grade tumors (Edmondson-Steiner grade I and II) had significantly lower mean intensity and higher GLRLM-nonuniformity values compared with high-grade tumors (grade III and IV). Lower mean intensity of high-grade tumors in this study is caused by intratumoral necrosis when the tumors advance and outgrow their blood supply – equating to reduced contrast delivery. GLRLM-nonuniformity denotes heterogeneity in tumor cellularity, vascularity, and mesenchymal distribution; hence the higher values in more aggressive tumors [76]. High-grade HCCs had significantly higher MPP (pixels with intensity values greater than 0) and SD compared with low-grade tumors [69]. In this study, however, the higher intensity values (indicating more contrast uptake, i.e. more blood flow) is probably due to angiogenesis, which is necessary for rapidly growing tumors to meet nutrient and oxygen requirements [77]. Perhaps intensity features change with growth and advancement of the tumor – with higher intensity values in rapidly growing tumors and lower values when they outgrow their blood supply and subsequently develop necrosis.

The current optimization strategies to realizing a standard HCC radiomics

Most studies on the texture quantification were conducted on brain, lung, or breast tumors. Similar studies on liver tumors are needed to assess further the robustness of texture features extracted from different types of HCC, as demonstrated by Perrin et al. [9]. Several scientific organizations are making efforts to standardize the processes of texture analysis to enable its full incorporation into clinical practice. Recently, the use of radiomics software tools that comply with IBSI’s standards (Imaging Biomarker Standardization Initiative), an initiative that seeks to standardize radiomics feature extraction, has been advocated [5]. Furthermore, radiomics analysis using a controlled imaging protocol will reduce the variability of extracted texture features [7]. To achieve the best evidence-based medical practice and enable a smooth clinical application of radiomics, a full disclosure of technical details and a clear description of the radiomic models and other processes of data analyses is necessary to enable external validation of results, to foster a better comparison of different study findings, and enable sharing of data as proposed by [24].

Our review focused on CT and MR-based HCC radiomics because most of the studies in HCC texture quantification were conducted using CT and MRI and the challenges associated with their radiomics process have been better identified and addressed compared to the few radiomics studies using other techniques such as multimodal ultrasound and positron emission tomography (PET). However, just like the BCLC algorithm, which incorporates tumor characteristics and the liver function status, studies using multimodal imaging techniques, such as the PET, to analyze both tumor and liver texture signatures are needed. This will help in developing a robust algorithm that will incorporate not only HCC radiomics, but also radiomics of the liver function.

Conclusion

Numerous studies have demonstrated the application of texture quantification, especially when combined with clinical and pathological variables, in the management of various forms and stages of HCC. However, the clinical validation of HCC radiomics has been limited by a lack of standardization in image acquisition protocol and optimization of the radiomics analysis procedure. Recent pioneering studies have identified some robust tumor radiomics signatures that are most resistant to protocol variations. Thus, further studies on these robust signatures and the development of a multiparameter-model that automatically corrects the discrepancies in the most influential data acquisition parameters or a comprehensive algorithm using a controlled imaging protocol can help reduce the heterogeneity in the quantification of tumor texture. Finally, the radiomics signature of tumor and liver parenchyma maybe useful for simultaneously assessing both tumor heterogeneity and residual liver function. Perhaps with further work, such functional radiomic signatures would prove useful in establishing a standard HCC radiomic algorithm.

Availability of data and materials

Not applicable.

Abbreviations

HCC:

Hepatocellular carcinoma

CT:

Computed tomography

MRI:

Magnetic resonance imaging

BCLC:

Barcelona Clinic Liver Cancer

SNR:

Signal-to-noise ratio

GLCM:

Gray-level co-occurrence matrix

GLRLM:

Gray-level run-length matrix

TE:

Echo time

TR:

Repetition time

SBW:

Sampling bandwidth

ROI:

Region of interest

VOI:

Volume of interest

MVI:

Microvascular invasion

ICC:

Intraclass correlation coefficient

ROC:

Receiver operating characteristic curve

AUC:

Area under the curve

References

  1. Chen WQ, et al. Cancer statistics in China, 2015. CA Cancer J Clin. 2016;66(2):115–32.

    PubMed  Google Scholar 

  2. Bruix J, et al. Clinical management of hepatocellular carcinoma. Conclusions of the Barcelona-2000 EASL conference. J Hepatol. 2001;35(3):421–30.

    CAS  PubMed  Google Scholar 

  3. Haralick RM, Shanmugam K, Dinstein IH. Textural features for image classification. IEEE Transactions on systems, man, and cybernetics. 1973;(6):610–21.

  4. Wakabayashi T, et al. Radiomics in hepatocellular carcinoma: a quantitative review. Hepatol Int. 2019;13(5):546–59.

    PubMed  Google Scholar 

  5. Lewis S, Hectors S, Taouli B. Radiomics of hepatocellular carcinoma. Abdom Radiol (NY). 2020;10.1007/s00261-019-02378-5. https://doi.org/10.1007/s00261-019-02378-5. [published online ahead of print, 2020 Jan 10].

  6. Miranda Magalhaes Santos JM, et al. State-of-the-art in radiomics of hepatocellular carcinoma: a review of basic principles, applications, and limitations. Abdom Radiol (NY). 2020;45(2):342–53.

    Google Scholar 

  7. Ger RB, et al. Comprehensive investigation on controlling for CT imaging Variabilities in Radiomics studies. Sci Rep. 2018;8:14.

    Google Scholar 

  8. Qiu Q, et al. Reproducibility and non-redundancy of radiomic features extracted from arterial phase CT scans in hepatocellular carcinoma patients: impact of tumor segmentation variability. Quant Imaging Med Surg. 2019;9(3):453–64.

    PubMed  PubMed Central  Google Scholar 

  9. Perrin T, et al. Short-term reproducibility of radiomic features in liver parenchyma and liver malignancies on contrast-enhanced CT imaging. Abdominal Radiol. 2018;43(12):3271–8.

    Google Scholar 

  10. Solomon J, et al. Quantitative features of liver lesions, lung nodules, and renal stones at multi-detector row CT examinations: dependency on radiation dose and reconstruction algorithm. Radiology. 2016;279(1):185–94.

    PubMed  Google Scholar 

  11. Zhao B, et al. Exploring variability in CT characterization of tumors: a preliminary phantom study. Transl Oncol. 2014;7(1):88–93.

    PubMed  PubMed Central  Google Scholar 

  12. Stocker D, et al. MRI texture analysis for differentiation of malignant and benign hepatocellular tumors in the non-cirrhotic liver. Heliyon. 2018;4(11):e00987.

    PubMed  PubMed Central  Google Scholar 

  13. Mackin D, et al. Effect of tube current on computed tomography radiomic features. Sci Rep. 2018;8(1):2354.

    PubMed  PubMed Central  Google Scholar 

  14. Kumar V, et al. Radiomics: the process and the challenges. Magn Reson Imaging. 2012;30(9):1234–48.

    PubMed  PubMed Central  Google Scholar 

  15. Mayerhoefer ME, et al. Effects of MRI acquisition parameter variations and protocol heterogeneity on the results of texture analysis and pattern discrimination: an application-oriented study. Med Phys. 2009;36(4):1236–43.

    PubMed  Google Scholar 

  16. Yang F, et al. Evaluation of radiomic texture feature error due to MRI acquisition and reconstruction: a simulation study utilizing ground truth. Phys Med. 2018;50:26–36.

    PubMed  Google Scholar 

  17. Ford J, et al. Quantitative Radiomics: impact of pulse sequence parameter selection on MRI-based textural features of the brain. Contrast Media Mol Imaging. 2018;2018:1729071.

    PubMed  PubMed Central  Google Scholar 

  18. Jirak D, Dezortova M, Hajek M. Phantoms for texture analysis of MR images. Long-term and multi-center study. Med Phys. 2004;31(3):616–22.

    PubMed  Google Scholar 

  19. Waugh SA, et al. Magnetic resonance imaging texture analysis classification of primary breast cancer. Eur Radiol. 2016;26(2):322–30.

    CAS  PubMed  Google Scholar 

  20. Savio SJ, et al. Effect of slice thickness on brain magnetic resonance image texture analysis. Biomed Eng Online. 2010;9:60.

    PubMed  PubMed Central  Google Scholar 

  21. Li J, Wang J, Lei L, Yuan G, He S. The diagnostic performance of gadoxetic acid disodium-enhanced magnetic resonance imaging and contrast-enhanced multi-detector computed tomography in detecting hepatocellular carcinoma: a meta-analysis of eight prospective studies. Eur Radiol. 2019;29(12):6519–28. https://doi.org/10.1007/s00330-019-06294-6.

  22. Zhang J, et al. Texture analysis based on preoperative magnetic resonance imaging (MRI) and conventional MRI features for predicting the early recurrence of single hepatocellular carcinoma after hepatectomy. Acad Radiol. 2019;26(9):1164–73.

    PubMed  Google Scholar 

  23. Mahon RN, Hugo GD, Weiss E. Repeatability of texture features derived from magnetic resonance and computed tomography imaging and use in predictive models for non-small cell lung cancer outcome. Phys Med Biol. 2019;64(14):14.

  24. Lambin P, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14(12):749–62.

    PubMed  Google Scholar 

  25. Castellano G, et al. Texture analysis of medical images. Clin Radiol. 2004;59(12):1061–9.

    CAS  PubMed  Google Scholar 

  26. Liu Z, et al. The applications of radiomics in precision diagnosis and treatment of oncology: opportunities and challenges. Theranostics. 2019;9(5):1303–22.

    PubMed  PubMed Central  Google Scholar 

  27. Ganeshan B, Miles KA. Quantifying tumour heterogeneity with CT. Cancer Imaging. 2013;13:140–9.

    PubMed  PubMed Central  Google Scholar 

  28. Limkin EJ, et al. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann Oncol. 2017;28(6):1191–206.

    CAS  PubMed  Google Scholar 

  29. Xu X, et al. Radiomic analysis of contrast-enhanced CT predicts microvascular invasion and outcome in hepatocellular carcinoma. J Hepatol. 2019;70(6):1133–44.

    PubMed  Google Scholar 

  30. Parmar C, et al. Robust Radiomics feature quantification using semiautomatic volumetric segmentation. PLoS One. 2014;9(7):e102107.

    PubMed  PubMed Central  Google Scholar 

  31. Heimann T, et al. Comparison and evaluation of methods for liver segmentation from CT datasets. IEEE Trans Med Imaging. 2009;28(8):1251–65.

    PubMed  Google Scholar 

  32. Yan J, Schwartz LH, Zhao B. Semiautomatic segmentation of liver metastases on volumetric CT images. Med Phys. 2015;42(11):6283–93.

    PubMed  PubMed Central  Google Scholar 

  33. Liao M, et al. Automatic liver segmentation from abdominal CT volumes using graph cuts and border marching. Comput Methods Prog Biomed. 2017;143:1–12.

    Google Scholar 

  34. Yang X, et al. A hybrid semi-automatic method for liver segmentation based on level-set methods using multiple seed points. Comput Methods Prog Biomed. 2014;113(1):69–79.

    Google Scholar 

  35. Li BN, et al. A new unified level set method for semi-automatic liver tumor segmentation on contrast-enhanced CT images. Expert Syst Appl. 2012;39(10):9661–8.

    Google Scholar 

  36. Yip SSF, et al. Associations between radiologist-defined semantic and automatically computed radiomic features in non-small cell lung cancer. Sci Rep. 2017;7(1):3519.

    PubMed  PubMed Central  Google Scholar 

  37. Segal E, et al. Decoding global gene expression programs in liver cancer by noninvasive imaging. Nat Biotechnol. 2007;25(6):675–80.

    CAS  PubMed  Google Scholar 

  38. Renzulli M, Brocchi S, Cucchetti A, et al. Can current preoperative imaging be used to detect microvascular invasion of hepatocellular carcinoma? Radiology. 2015;279(2):432–42. https://doi.org/10.1148/radiol.2015150998.

  39. Banerjee S, et al. A computed tomography radiogenomic biomarker predicts microvascular invasion and clinical outcomes in hepatocellular carcinoma. Hepatology. 2015;62(3):792–800.

    PubMed  PubMed Central  Google Scholar 

  40. Echegaray S, et al. Core samples for radiomics features that are insensitive to tumor segmentation: method and pilot study using CT images of hepatocellular carcinoma. J Med Imaging (Bellingham). 2015;2(4):041011.

    Google Scholar 

  41. Yang L, et al. A radiomics nomogram for preoperative prediction of microvascular invasion in hepatocellular carcinoma. Liver Cancer. 2019;8(5):373–86.

    CAS  PubMed  Google Scholar 

  42. Kim J, et al. Predicting survival using pretreatment CT for patients with hepatocellular carcinoma treated with Transarterial chemoembolization: comparison of models using radiomics. AJR Am J Roentgenol. 2018;211(5):1026–34.

    PubMed  Google Scholar 

  43. Sanghera B, et al. Reproducibility of 2D and 3D fractal analysis techniques for the assessment of spatial heterogeneity of regional blood flow in rectal cancer. Radiology. 2012;263(3):865–73.

    PubMed  Google Scholar 

  44. Al-Kadi OS. Assessment of texture measures susceptibility to noise in conventional and contrast enhanced computed tomography lung tumour images. Comput Med Imaging Graph. 2010;34(6):494–503.

    PubMed  Google Scholar 

  45. El Naqa I, et al. Exploring feature-based approaches in PET images for predicting cancer treatment outcomes. Pattern Recogn. 2009;42(6):1162–71.

    Google Scholar 

  46. Davnall F, et al. Assessment of tumor heterogeneity: an emerging imaging tool for clinical practice? Insights Imaging. 2012;3(6):573–89.

    PubMed  PubMed Central  Google Scholar 

  47. Blanc-Durand P, et al. Signature of survival: a (18) F-FDG PET based whole-liver radiomic analysis predicts survival after (90) Y-TARE for hepatocellular carcinoma. Oncotarget. 2018;9(4):4549–58.

    PubMed  Google Scholar 

  48. Ng F, et al. Assessment of tumor heterogeneity by CT texture analysis: can the largest cross-sectional area be used as an alternative to whole tumor analysis? Eur J Radiol. 2013;82(2):342–8.

    PubMed  Google Scholar 

  49. Kim S, et al. Radiomics on gadoxetic acid–enhanced magnetic resonance imaging for prediction of postoperative early and late recurrence of single hepatocellular carcinoma. Clin Cancer Res. 2019;25(13):3847–55.

    CAS  PubMed  Google Scholar 

  50. Ahn SY, et al. Prediction of microvascular invasion of hepatocellular carcinoma using gadoxetic acid-enhanced MR and (18) F-FDG PET/CT. Abdom Imaging. 2015;40(4):843–51.

    PubMed  Google Scholar 

  51. Jain AK, Duin RPW, Mao JC. Statistical pattern recognition: a review. IEEE Trans Pattern Anal Mach Intell. 2000;22(1):4–37.

    Google Scholar 

  52. Bellman RE. Adaptive control processes: a guided tour, vol. 2045. New Jersey: Princeton university press; 2015.

  53. Bagherzadeh-Khiabani F, et al. A tutorial on variable selection for clinical prediction models: feature selection methods in data mining could improve the results. J Clin Epidemiol. 2016;71:76–85.

    PubMed  Google Scholar 

  54. Miles KA, Ganeshan B, Hayball MP. CT texture analysis using the filtration-histogram method: what do the measurements mean? Cancer Imaging. 2013;13(3):400–6.

    PubMed  PubMed Central  Google Scholar 

  55. Ni M, et al. Radiomics models for diagnosing microvascular invasion in hepatocellular carcinoma: which model is the best model? Cancer Imaging. 2019;19(1):60.

    PubMed  PubMed Central  Google Scholar 

  56. Ma X, et al. Preoperative radiomics nomogram for microvascular invasion prediction in hepatocellular carcinoma using contrast-enhanced CT. Eur Radiol. 2019;29(7):3595–605.

    PubMed  Google Scholar 

  57. Zhang R, et al. A nomogram based on bi-regional radiomics features from multimodal magnetic resonance imaging for preoperative prediction of microvascular invasion in hepatocellular carcinoma. Quant Imaging Med Surg. 2019;9(9):1503–15.

    PubMed  PubMed Central  Google Scholar 

  58. Zhu YJ, et al. Model-based three-dimensional texture analysis of contrast-enhanced magnetic resonance imaging as a potential tool for preoperative prediction of microvascular invasion in hepatocellular carcinoma. Oncol Lett. 2019;18(1):720–32.

    PubMed  PubMed Central  Google Scholar 

  59. Zhang Z, et al. Hepatocellular carcinoma: radiomics nomogram on gadoxetic acid-enhanced MR imaging for early postoperative recurrence prediction. Cancer Imaging. 2019;19(1):22.

    CAS  PubMed  PubMed Central  Google Scholar 

  60. Zhou Y, et al. CT-based radiomics signature: a potential biomarker for preoperative prediction of early recurrence in hepatocellular carcinoma. Abdominal Radiology. 2017;42(6):1695–704.

    PubMed  Google Scholar 

  61. Campos JT, Sirlin CB, Choi J-Y. Focal hepatic lesions in Gd-EOB-DTPA enhanced MRI: the atlas. Insights into Imaging. 2012;3(5):451–74.

    PubMed  PubMed Central  Google Scholar 

  62. Li Z, et al. Texture-based classification of different single liver lesion based on SPAIR T2W MRI images. BMC Med Imaging. 2017;17(1):42.

    PubMed  PubMed Central  Google Scholar 

  63. Fu S, et al. Texture analysis of intermediate-advanced hepatocellular carcinoma: prognosis and patients' selection of transcatheter arterial chemoembolization and sorafenib. Oncotarget. 2017;8(23):37855–65.

    PubMed  Google Scholar 

  64. Li M, et al. Computed tomography texture analysis to facilitate therapeutic decision making in hepatocellular carcinoma. Oncotarget. 2016;7(11):13248–59.

    PubMed  PubMed Central  Google Scholar 

  65. Kloth C, et al. Evaluation of texture analysis parameter for response prediction in patients with hepatocellular carcinoma undergoing drug-eluting bead transarterial chemoembolization (DEB-TACE) using biphasic contrast-enhanced CT image data: correlation with liver perfusion CT. Acad Radiol. 2017;24(11):1352–63.

    PubMed  Google Scholar 

  66. Yu JY, et al. Value of texture analysis based on enhanced MRI for predicting an early therapeutic response to transcatheter arterial chemoembolisation combined with high-intensity focused ultrasound treatment in hepatocellular carcinoma. Clin Radiol. 2018;73(8):10.

    Google Scholar 

  67. Kiryu S, et al. Impact of hepatocellular carcinoma heterogeneity on computed tomography as a prognostic indicator. Sci Rep. 2017;7(1):12689.

    PubMed  PubMed Central  Google Scholar 

  68. Mule S, et al. Advanced hepatocellular carcinoma: pretreatment contrast-enhanced CT texture parameters as predictive biomarkers of survival in patients treated with Sorafenib. Radiology. 2018;288(2):445–55.

    PubMed  Google Scholar 

  69. Oh J, et al. Hepatocellular carcinoma: texture analysis of preoperative computed tomography images can provide markers of tumor grade and disease-free survival. Korean J Radiol. 2019;20(4):569–79.

    PubMed  PubMed Central  Google Scholar 

  70. Song S, Li Z, Niu L, et al. Hypervascular hepatic focal lesions on dynamic contrast-enhanced CT: preliminary data from arterial phase scans texture analysis for classification. Clin Radiol. 2019;74(8):653.e11-653.e18. https://doi.org/10.1016/j.crad.2019.05.010.

  71. Mokrane FZ, et al. Radiomics machine-learning signature for diagnosis of hepatocellular carcinoma in cirrhotic patients with indeterminate liver nodules. Eur Radiol. 2020;30(1):558–70.

    PubMed  Google Scholar 

  72. Huang YL, Chen JH, Shen WC. Diagnosis of hepatic tumors with texture analysis in nonenhanced computed tomography images. Acad Radiol. 2006;13(6):713–20.

    PubMed  Google Scholar 

  73. Brenet Defour L, et al. Hepatocellular carcinoma: CT texture analysis as a predictor of survival after surgical resection. Eur Radiol. 2019;29(3):1231–9.

    PubMed  Google Scholar 

  74. Zheng BH, et al. Radiomics score: a potential prognostic imaging feature for postoperative survival of solitary HCC patients. BMC Cancer. 2018;18(1):1148.

    PubMed  PubMed Central  Google Scholar 

  75. Song W, Yu X, Guo D, et al. MRI-Based Radiomics: Associations With the Recurrence-Free Survival of Patients With Hepatocellular Carcinoma Treated With Conventional Transcatheter Arterial Chemoembolization. J Magn Reson Imaging. 2020;52(2):461–73. https://doi.org/10.1002/jmri.26977.

  76. Zhou W, et al. Malignancy characterization of hepatocellular carcinomas based on texture analysis of contrast-enhanced MR images. J Magn Reson Imaging. 2017;45(5):1476–84.

    PubMed  Google Scholar 

  77. Park YN, et al. Increased expression of vascular endothelial growth factor and angiogenesis in the early stage of multistep hepatocarcinogenesis. Arch Pathol Lab Med. 2000;124(7):1061–5.

    CAS  PubMed  Google Scholar 

Download references

Acknowledgements

Not applicable.

Funding

This work has been supported by the Hunan Science and Technology Department Plan Project (grant number: 2018XK2304, Changsha, China) Natural Science Foundation of Hunan Province (grant number: 2018JJ2656, Changsha, China), China Postdoctoral Science Foundation(grant number: 2019 M652807, Beijing, China).

Author information

Authors and Affiliations

Authors

Contributions

IM and YP were the major contributors to the writing of the manuscript, WG, SX, and DM contributed to the literature search, and the writing of the manuscript, WL contributed to the critical revision of the article. All authors have read and approved the final manuscript.

Corresponding authors

Correspondence to Yigang Pei or Wenzheng Li.

Ethics declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Masokano, I.B., Liu, W., Xie, S. et al. The application of texture quantification in hepatocellular carcinoma using CT and MRI: a review of perspectives and challenges. Cancer Imaging 20, 67 (2020). https://doi.org/10.1186/s40644-020-00341-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s40644-020-00341-y

Keywords