Texture Analysis Methodology
Texture analysis methodology has varied across different studies with a wide spectrum of extracted texture features, segmentation techniques and software. In this section we will review the definition of most common texture features and methods of recently published studies.
Texture Features
The number and type of texture features may vary depending on the software and on the number of phases/sequences. Texture features may be divided in three main categories including first, second and third other features. The first order features are calculated from the analysis of the gray level histogram within a defined ROI, without considering spatial relations among pixels, including:
- Mean: corresponding to average value of the pixels with the region of interest.
- Standard deviation: variation of the pixels from the mean.
- Mpp: mean of the pixels with positive values.
- Skewness: reflects the asymmetry of the histogram.
- Kurtosis: reflects the peakedness or flatness of the histogram.
- Entropy: quantifies image irregularity or complexity.
The second order texture features consider the spatial relations among pixels, and most commonly include:
- Grey level co-occurrence matrix (GLCM): quantifies the arrangements of pairs of pixels with the same values in specific directions.
- Grey-level run length matrix (GLRLM): quantifies consecutive pixels with the same intensity along specific directions.
Third or higher order features evaluate spatial relationship among three or more pixels through statistical methods after applying filters or mathematical transforms. These features include fractal analysis, wavelet transform, and Laplacian transforms of Gaussian-filtered image.
Feature Extraction
Imaging Methods
Texture analysis may be potentially applied on any imaging modalities, including ultrasound, CT or MRI (Fig. 2). Although both non-contrast and contrast-enhanced images may be evaluated, the latter may suffer from heterogeneous contrast timing, type of contrast agent, and injection rate. Moreover, most of the texture features are highly affected by the acquisition and scanning parameters, with reconstruction thickness having one of the highest impact on features reproducibility.
Fig. 2: Imaging modalities in recent texture analysis studies [1-13].
References: Section of Radiology, BIND, University of Palermo, Palermo/IT
CT studies have used most commonly the portal venous phase for texture analysis (Fig. 3) since it is included in all imaging protocols. As opposed, MRI studies have selected most frequently the T1-weigthed pre-contrast images (Fig. 4).
Fig. 3: Most common CT phases selected for texture analysis studies [1-13]. Pre: pre-contrast images; HAP: hepatic arterial phase; PVP: portal venous phase; DP: delayed phase.
References: Section of Radiology, BIND, University of Palermo, Palermo/IT
Fig. 4: Most common MRI sequences selected for texture analysis studies [1-13]. T1 Pre: T1-weighetd images before contrast administration; T2: T2-weighted sequences; HAP: hepatic arterial phase; PVP: portal venous phase; DP: delayed phase; HBP: hepatobiliary phase; CCE: combined-contrast-enhanced.
References: Section of Radiology, BIND, University of Palermo, Palermo/IT
Segmentation Techniques
The segmentation represents a critical step for texture analysis studies. Quantitative data are extracted by placing ROI in the hepatic parenchyma. Too small or too large ROI may cause undersampling or include potential confounding areas such as adjacent organs, hepatic vessels, bile ducts or liver lesions.
Segmentation may be performed manually, with semi-automatic or automatic techniques. Manual segmentation by experienced radiologist is considered the gold standard, but it is time-consuming and may suffer of intra- and inter-reader variability. Automatic segmentation may be more objective, but it is prone to errors in case of imaging artifacts or liver lesions. Segmentation may be performed with 2D ROI, when placed on a single slide, or 3D ROI, when includes a volume of liver parenchyma in multiple consecutive images or the whole liver. Although 3D analysis may capture more information, some studies [3, 8, 9] have demonstrated that single-slice analysis is often sufficient for the evaluation of hepatic fibrosis. The most common segmentation techniques include (Fig. 5):
- Single ROI placed on a specific hepatic segment, not including large hepatic vessels or focal liver lesions.
- Multiple ROI placed on different hepatic segments or in different levels, not including large hepatic vessels or focal liver lesions.
- ROI including the whole liver parenchyma or specific segments, placed at the level of the porta hepatis, without including major hepatic vessels or biliary ducts.
Fig. 5: Most common segmentation methods adopted in texture analysis studies for evaluation of hepatic fibrosis.
References: Section of Radiology, BIND, University of Palermo, Palermo/IT
Texture Analysis Software
Software for texture analysis include:
- Commercially available software.
- Freely available online software (MaZda, LIFEx).
Nowadays, the variability of software and tools for texture analysis make the comparison among different studies challenging.
Current Evidences
The following Table summarizes the most recent texture analysis studies including number of patients, imaging technique and diagnostic performance of most significant texture features or statistical models including combination of multiple features for the detection of advanced fibrosis-cirrhosis.
Table 1: Most recent studies including number of patients, etiology of the chronic liver disease (CLD), imaging technique and diagnostic performance based on the reported area under the ROC curve (AUROC) results.
References: Section of Radiology, BIND, University of Palermo, Palermo/IT
Most studies have included patients with hepatitis C infection (HCV) or hepatitis B infection as etiology of the chronic liver disease (59% and 23%, respectively). Interesting, only few patients (4% of all subjects included in recent studies) had non-alcoholic fatty liver disease (NAFLD), which is becoming the main etiology of cirrhosis in Western Countries (Fig. 6).
Fig. 6: Distribution of patients included in recently published texture analysis studies [1-13].
References: Section of Radiology, BIND, University of Palermo, Palermo/IT
Strengths and Limitations of Texture Analysis
Strengths and weaknesses of texture analysis and other non-invasive imaging methods for the assessment of hepatic fibrosis are summarized in the following Table.
Table 2: Strengths and weaknesses of non-invasive methods for the assessment of hepatic fibrosis.
References: Section of Radiology, BIND, University of Palermo, Palermo/IT
Future perspectives
Further studies are needed to assess several gaps knowledge:
- Evaluation of large population of NAFLD or other etiologies of CLD such as primary biliary cirrhosis.
- Validation in multicentric studies with different patients’ population and imaging parameters.
- Evaluation of intra- and inter-reader agreement.
- Comparison with other non-invasive techniques for fibrosis evaluation.