HCC was a rapidly growing tumor with more than 500,000 new cases per year[6], and the recurrence rate within the Milan criteria (ie, a single HCC tumor <5 cm or all three tumors >3 cm) was gradually increasing[7]. Rou WS[8] studied 134 patients initially diagnosed with HCC and identified factors affecting early local recurrence after TACE. They found that tumor size greater than 2cm significantly predicted early local recurrence after complete TACE. Similar to our results, there was a significant correlation between local recurrence and tumor size for the entire cohort. However, tumor size alone cannot predict response to treatment in early locally recurrent lesions due to the relative resistance of existing undifferentiated cells in TACE-induced hypoxic condition.
Su X et al[9] discussed that most treatments were not adequately addressed due to tumor heterogeneity, which may be one of the most important factors. Tumor heterogeneity represents hemorrhage, necrosis, and regional differences in cell density[10]. Not only can it be visualized , but also it manifests at the cellular and molecular levels due to the morphological diversity[11]. Intratumor heterogeneity, as the most promising prognostic factor, combined with histopathologic grade can predict survival[12] and therapeutic response[13].
On contrast-enhanced CT imaging, larger HCCs lead to markedly heterogeneous enhancement due to structural inhomogeneity. Traditionally, CT values typically reflect average density within a defined region of interest (ROI) and may not be suitable for assessment of tumor biological behavior before treatment. Fortunately, histogram and texture analysis, currently the most popular techniques, have been applied in medical images to quantify the heterogeneity of the whole tumor. CT-enhanced imaging-based histogram measurements, such as percentile values, minimum, and maximum, can account for potential heterogeneous distributions and thus be used to predict treatment response[14] and correlate with pathological outcomes[15]. Texture analysis provides information about the tumor microenvironment through a variety of mathematical methods that consider not only the overall pixel intensity, but the location and distribution of pixel pairs in the image[16-17]. It can quantify and characterize heterogeneous distributions by providing better representations and has been successfully applied in several areas, including distinguishing benign and malignant lesions[18] and the prediction of survival time[19]. However, to date, the application of histogram and texture analysis to predict HCC tumor recurrence is not very widespread.
According to recent research, heterogeneity is the dominating factor in cancer recurrence[20-21]. It may be attributed to a heterogeneous tumor cell population composed of cancer stem cells that had been isolated and characterized in many types of cancers[22]. Previous studies have shown that the extracellular matrix (ECM) is an important part of the tumor microenvironment, consisting of a variety of stromal cells, which play an important role in tumor recurrence and progression by interacting with cancer cells or with each other. Intratumoral heterogeneity of HCC exists not only at the morphological but molecular level[7], including immunehistochemical and genetic intratumor heterogeneity. Through a systematic analysis of 23 HCC without medical pretreatment, Friemel et al[23] found that morphological heterogeneity was detectable in 87% of HCC cases, and immunohistochemical heterogeneity was consistently associated with morphological abnormalities in 39% of tumors. These findings might be responsible for treatment failure in many HCC cases.
Tumor vascularity was assessed by analyzing data from preprocessed functional imaging, supported by a number of previous studies. Histogram and texture characteristics have been shown to associate with clinical outcomes such as response to therapy in a variety of tumors [16][24]. Reiner CS et al [16] measured arterial perfusion to assess whether tumor heterogeneity helps predict response to transarterial radioembolization (TARE) by histogram analysis of CT perfusion. They found that responders had significantly higher arterial perfusion than non-responders, suggesting that higher vascularization was associated with better TARE responses[25]. Selective catheterization is very important to achieve effective TACE therapy in patients with hypervascular HCC. Similar to Reiner CS’s study[16], according to our histogram results, the non-ER group had significantly higher pixel intensities at the 10th and 50th percentiles.
In fact, in a study by Chang Y et al[26], the lifetime of HCC patients with vascularity was significantly longer than that of oligovascular patients (P<0.01) for TACE treatment. This was similar to our results, which showed a strong association between tumor heterogeneity using histogram and texture analysis and response to TACE in HCC patients. Contrast-enhanced CT has been used in a number of clinical trials to assess and monitor the effect of therapy. However, influenced by tumor microenvironment[27], the average CT value drawn on a single slice could not favorably reflect the heterogeneity in the whole volume tumor. Retrospective analysis of histograms and textures using standard portal-phase liver CT scans not only did not affect survival prediction[28], but eliminated concerns that data from these studies could confound extreme results. Furthermore, the choice of interest is another important influencing factor, considering intratumoral heterogeneity. Our texture evaluation results were very similar when comparing ROI with VOI mode. Previous studies described heterogeneous disease characterization based on whole lesions rather than single slice analysis, which could reduce selection bias [14][29]. Consistently, Ng F et al[30] found that whole-tumor analysis was more representative of tumor heterogeneity than maximum cross-sectional area analysis. Given the above, our research analysis was conducted within the VOI.
Texture analysis, an image processing algorithm, can be used to assess the pixel distribution within the tumor to quantify texture. Entropy is an important parameter for statistical measurement of gray intensity irregularity, and the entropy tends to be higher when there are more variable values in the region of interest. Entropy and inhomogeneity strongly reflect heterogeneity in tumor characterization[31] related to tissue density, angiogenesis and fibrosis[32]. Our study found that entropy and inhomogeneity of the ER group were positively correlated with early recurrence, suggesting that tumor heterogeneity in HCC correlated negatively with response to TACE. The lower pixel value of ER group supported the argument. Skewness and kurtosis describe the asymmetry and the sharpness of the distribution of data points compared to the normal Gaussian distribution. They are more sensitive to outliers. In Lubner MG`s study [33], skewness was negatively associated with KRAS mutation, but there was not a significant predictive value in our analysis and prior study [24]. Similar to Hounsfield unit (HU), the mean mainly reflects the average vascular permeability measured on contrast-enhanced CT imaging.
There were several limitations in our study. First, our research analyzed the whole tumor volume heterogeneity of the target lesion, instead of taking into account the spatial distribution within the tumor. Firstly, intralesional heterogeneity in HCC has important implications for targeted therapy, suggesting that a single tumor biopsy is not representative of the entire tumor [23], which may result in an incomplete therapy response due to the sensitivity of only one tumor subclone. Secondly, Although our results suggested that histogram and texture analysis were able to predict response to treatment, manual drawing of lesion boundaries might be biased. Improving the algorithm for edge detection and developing a fully automated method will be the next task to improve reproducibility. Thirdly, TACE is considered a selective method used in the terminal stage of HCC. Many patients in poor health could not undergo long-term follow-up imaging. This excluded many patients originally and became the handicap to increase the patients’ database. Tumor heterogeneity indicates differences in spatial distribution, including hypoxia, necrosis, tumor solid area, and peritumoral edema, which may further illustrate the reactivating lesion and lucubrately need more research.