Potential of Using qFibrosis Analysis to Predict Recurrent and Survival Outcome of Patients with Hepatocellular Carcinoma after Hepatic Resection

Hepatocellular carcinoma (HCC) has always been a great threat to the people’s health globally. Surgical resection remains the best choice for radical treatment, with the lowest recurrence rate and better survival outcome. Recurrence of HCC is an important issue affecting the long-term survival of patients. Even after surgical resection, the average recurrence rate in 5 years is still more than 60% [1].

Many prognostic factors for HCC had been reported. The important independent prognostic factors include tumor stage (size, number, vascular invasion), serum alpha fetoprotein level, and surgical resection margin [2‒6]. However, there is still a lack of studies using patients’ histopathological features as a parameter to predict patient outcome. This was because of the lack of suitable tools to quantitatively assess patients’ tissue effectively and efficiently in the past. In recent years, the popularization of artificial intelligence and the advancement of algorithms have brought attention to the quantitative analysis of fibrosis, i.e., qFibrosis. qFibrosis is a system of second harmonic generation (SHG) and two-photon excitation fluorescence (TPEF) microscopy which can identify, quantify, and visualize the fibrosis features in biopsy samples. qFibrosis has been successfully validated for its application in the acute diagnosis of fibrosis and the prognosis of hepatitis B [7‒10] and non-alcoholic steatohepatitis [11‒16].

Fibrosis is characterized by excessive accumulation of collagen and other extracellular matrix components, and this process could be seen in chronic liver disease and cancer development. The extracellular matrix maintains normal tissue polarity and architecture and also has a role in preventing cancer cell invasion [17‒22]. It is an essential component of the stem cell and cancer stem cell niche. Deregulation and remodeling of the extracellular matrix are key processes of cancer development. Recent studies have revealed the important role of stromal remodeling and how stromal fibroblasts in the tumor microenvironment can promote cancer initiation and progression [17, 22‒24]. Stromal remodeling is featured by collagen realignment in the stromal compartment, collagen fibers, and basal membrane, which contribute to the fibrosis features in both tumor and non-tumor part of liver tissue. However, there remains a lack of studies addressing the stromal background and fibrosis features and their prognostic value in liver cancer.

Whether histopathological features of patients with HCC could be used as a parameter to predict disease and survival outcomes has little been studied. We propose that the stromal background and fibrosis features analyzed using qFibrosis system in patients with HCC have predictive values for disease and patient outcome. In this study, we aimed to demonstrate the significance of collagen features in patients with HCC after curative hepatic resection.

This study was approved by the Ethics Committee and the Institute Review Board of National Taiwan University Hospital (Protocol ID: 201912069RINB). Human tissue sampling was conducted on stored surgical specimens obtained from 201 patients with HCC who underwent curative tumor resection at the National Taiwan University Hospital from February 2011 to January 2013. Written informed consent had been obtained from all patients before sample collection. The patients’ baseline characteristics, along with perioperative biochemical data and tumor pathology, and disease outcome as well as survival outcome were collected through the electronic medical records.

Patients received regular follow-up at the outpatient clinic after tumor resection. Tumor markers and routine blood tests for liver function were checked at each visit, and an abdominal imaging examination with sonography, computed tomography, or magnetic resonance imaging was performed every 3–6 months, or as needed. Recurrence was established according to typical dynamic imaging or histopathologic confirmation. Early recurrence was defined by recurrence within 2 years after tumor resection. Early death was defined by death within 5 years after tumor resection.

Pathological specimens were fixed in phosphate-buffered neutral formalin, embedded in paraffin, and sectioned using standard clinical techniques. Five-micrometer sections were stained with hematoxylin and eosin, reticulin, and Masson’s trichrome for histological assessment. We profiled these 201 liver cancer specimens as well as their normal part of liver tissue with SHG/TPEF parameters and correlated them with the patient’s tumor status of portal vein thrombosis, recurrence-free survival (RFS), and overall survival (OS). A total of 201 HCC patient specimens will be studied using SHG/TPEF microscopy. The scanning was performed within the tumor (tumor part) and outside the tumor (liver tissue of non-tumor part). The scanning area was set at 2 mm × 2 mm with spatial resolution of 0.2 micrometer for these images. After scanning all specimens, the images were stored and processed for quantified and qualified analysis. Local morphological features of ECM, defined as different collagen parameters, will be extracted from SHG/TPEF images and correlated with clinicopathological data and long-term clinical outcomes (over survival time and RFS time) to identify potential prognostic collagen parameters.

The five-micrometer-thick sections of 201 unstained non-tumor liver samples and 201 unstained tumor samples were imaged using the Genesis system (HistoIndex Pte. Ltd., Singapore), in which SHG microscopy serves to visualize collagen and TPEF microscopy highlights hepatocytes. The samples were laser-excited at 780 nm, SHG signals were recorded at 390 nm, and TPEF signals were recorded at 550 nm. Images were acquired at ×20 magnification with ×512 512 pixels of resolution, and each image represented a 200 × 200 μm² area of tissue. Multiple adjacent images were captured to encompass a tissue region of 24 mm² (6 × 4 mm).

The SHG/TPEF images of unstained slides were analyzed by a digital image processing algorithm. For the non-tumor part, the algorithm can detect the various regions in liver tissue, including the portal tract, periportal, zone 2, pericentral, and central vein regions. And the fibrosis parameters, such as the collagen proportionate area and the number of collagen strings, can be quantified in each region. A total 156 fibrosis parameters were quantified in the non-tumor part. For the tumor part, the algorithm can detect a total of 33 fibrosis parameters in the overall region.

Two prediction models were developed based on the fibrosis parameters of tumor and non-tumor samples, including (1) model A: RFS index was used to predict early recurrence (RFS <2 years) and (2) model B: OS index was used to predict early death (OS <5 years). The constructions for the two models had similar processes, including (1) feature selection, (2) training based on the corresponding data, and (3) leave-one-out cross-validation. For example, the RFS index was used to predict early recurrence based on the 33 fibrosis parameters in the tumor part and the 156 fibrosis parameters in the non-tumor part. The samples were divided into two groups, i.e., the “early recurrence” group and the “no early recurrence” group. The sequential feature selection method was used to select 10 parameters to train the RFS index [25]. In the process of sequential feature selection, a linear regression model was employed. The selection criterion was the residual sum of squares, and the search algorithm utilized was sequential forward selection. The cut-off value was determined by Youden’s index. The model was validated by the leave-one-out method. The OS index used to predict early death was calculated throughout a similar process by dividing patient samples into two groups, i.e., the “early death” group and the “no early death” group, according to patients’ survival outcomes. Image quantification and model construction were performed using Matlab 2021a (The MathWorks, Inc., Natick, MA, USA). The illustration of sample scanning and quantification, along with examples of SHG/TPEF images displaying diverse collagen patterns in patients with different outcomes, is presented in Figure 1.

Data are expressed by the mean ± standard deviation, median (interquartile range), or number (percentage), as appropriate. The Wilcoxon rank sum test was performed to estimate the statistical difference in model indexes between the early recurrence group and the no early recurrence group, as well as the early death group and the no early death group. Disease-free curves and survival curves were calculated using the Kaplan-Meier method, and distributions were compared using the log-rank test. Statistical significance level was set at p < 0.05. All statistical analysis was performed using the R 4.2.2 software (R Foundation for Statistical Computing, Vienna, Austria).

The demographic data of the 201 patients are summarized in Table 1. The mean age was 58.7 ± 14 (range, 14–87) years. Male sex accounts for 79.1%. Most patients had good liver reserve with Child-Pugh score A (93%) or albumin-bilirubin (ALBI) grade 1 (95%). The median level of preoperative alpha fetoprotein was 19.9 (IQR, 3.9–337.9) ng/mL. The majority of patients had AJCC stage 1 disease (48.3%), followed by stage 2 (37.3%) disease. The median tumor size was 4.0 cm (range, 1.3–20.4 cm). Most patients had grade 2 (49.3%) and grade 3 (45.3%) tumor differentiation, with 25 (12.4%) patients present with microvascular invasion in the pathology. The median follow-up period was 85 months (IQR, 51.4–105.1 months), and all patients had been followed up for at least 2 years or until death. During follow-ups, 127 patients had recurrence, the median time of recurrence was 16.3 months (IQR: 7.4–38.4 months). Among patients with recurrence, 77 patients had early recurrence within 2 years. A total of 46 patients had early death within 5 years, and 143 patients survived for longer than 5 years. The median time of death was 30.2 months (range: 4.8–85.6 months).

A total of 10 fibrosis parameters (3 parameters from the tumor part and 7 parameters from the non-tumor part, Table 2) were selected to build the RFS index. The RFS index differentiates the patients with RFS ≥2 years (n = 124) and RFS <2 years (n = 77) with a cut-off value RFS index = 0.52, p < 0.001. Disease-free probability was lower in the high-risk group (RFS index <0.52) than in the low-risk group (RFS index ≥0.52), and the p value was <0.001 (Fig. 2). Collagen feature of a specimen with a higher value of the RFS index appears to have a lower risk to have early recurrence. AJCC cancer staging also showed a prediction value for early recurrence; however, tumor grade showed low correlation (Fig. 3).

A total of 10 fibrosis parameters (3 parameters from the tumor part and 7 parameters from the non-tumor part, Table 3) were selected to build the OS index. The OS index differentiates patients with early death (n = 46) and patients who survived longer than 5 years (n = 143), with a cut-off value of OS index = 0.73 (p = 0.02). The survival probability was lower in the high-risk group (OS index <0.73) than in the low-risk group (OS index ≥0.73), and the p value was <0.001 (Fig. 4). Collagen feature of a specimen sample with a higher value of the OS index has a lower risk to have early death. Both AJCC cancer staging and tumor grade showed good prediction performance for OS (Fig. 5).

Our study indicates that histological features of patients with HCC analyzed by qFibrosis appear to have prediction values for disease and survival outcomes. The combined index built using qFibrosis may have the ability to differentiate between early recurrence and early death in patients with HCC.

The application of artificial intelligence is getting more and more attention in the era of big data and precision medicine. Physicians use various information, including but not limited to biochemical data, tumor characteristics (such as size, number, staging), and genetic data, to assist their clinical decisions. All the information from patient may be collected as a reference for comprehensively judging the patient’s condition, treatment, and prognosis. Not surprisingly, cancer staging and tumor differentiation in our study had more or less some prediction value about patient’s disease and survival outcome; however, the value of histological characteristics of the patients had not yet received enough attention and research.

In the past decades, information on histopathological features of patients has mainly depended on the interpretation of a pathologist using traditional staining techniques. As a non-staining and quantitative analyzing technique, qFibrosis uses digital microscopy with artificial intelligence to provide information on histological features objectively and with high efficiency. qFibrosis has shown its usefulness in accurate fibrotic scoring of hepatic tissue in animal models and chronic hepatitis B patients [7‒10]. Besides, it had been applied to patients with chronic liver disease such as non-alcoholic steatohepatitis, in several clinical trials [11‒16]. However, the value of qFibrosis in clinical outcome prediction is still with little evidence.

Our study suggests that the application of qFibrosis might be correlated with the clinical outcome of HCC patients. The importance of this finding is that, like patients’ biochemical tests and genetic data, quantitative analysis of histological features could be regarded as one of the unique and personalized biological features, and this biological feature has the potential to be used to predict the prognosis of patients. Similar to our study, Lui et al. [26] have reported the positive results of using qFibrosis to predict the early recurrence of patients with HCC. Our study had a much larger (i.e., 3-fold) sample size with a much longer follow-up period, and we further showed the prediction value of this novel tool as for the survival outcome of HCC patients. The findings provide a new potential clinical application of this novel tool in the era of artificial intelligence and precision medicine.

Liver fibrosis could be modified by treatments such as antiviral therapy. In our country, effective antiviral drugs for the hepatitis B virus have been available and covered by national health insurance for over a decade. Consequently, the majority of patients in this study commenced antiviral treatment for hepatitis B immediately after their initial diagnosis of HCC. Conversely, most patients did not receive antiviral therapy for the hepatitis C virus, as effective treatments for it have only been accessible in recent years. As a result, our patient cohort demonstrates homogeneity regarding the clinical management of viral hepatitis. However, this homogeneity also indicates the absence of a sufficient control group for comparing fibrosis differences between treated and untreated patients. Investigating the impact of antiviral treatment on qFibrosis, both pre- and post-therapy, will be a primary focus of our future research endeavors.

Our study was limited by its retrospective designs, and the patient group consisted only of surgically treated patients. All patients included in this study had liver specimens obtained during tumor resection, typically representing relatively early-stage HCC patients who were candidates for curative treatment. Consequently, the predictive model developed in this study is applicable only to patients with pathology specimens containing both tumor and non-tumor parts of the liver. It cannot be extrapolated to patients with unresectable or distant metastasis-advanced HCC. This limitation may impact the generalizability of our findings. Prospective study with more patient number and including more generalized patient population would be helpful to further examine the powerfulness of the collagen feature analyzed by qFibrosis. The clinical relevance of the selected fibrosis parameters in the combined index and their pathological implications also need further investigation in future studies.

In conclusion, our study showed that the combined index calculated with qFibrosis from a digital readout of the fibrotic status of peri-tumor liver tissue in patients with HCC has prediction values for their disease and survival outcomes. The wider use of artificial intelligence in digital pathology helps to transform individual histopathological features into exploitable biological information which provides valuable potential concerning patient care in the future.

The research was conducted ethically in accordance with the World Medical Association Declaration of Helsinki. The study protocol was reviewed and approved and has been granted an exemption from requiring written informed consent by the Research Ethics Committee of National Taiwan University Hospital (NTUH REC: 201912069RINB).

The authors have no conflicts of interest to declare.

No funding was received in support of this work.

C.-Y.H., Y.R., and K.-W.H. performed most of the study and drafted the manuscript. E.C. and D.T. participated in the acquisition and statistical analysis of the data. D.T. and K.-W.H. contributed to the conception and design of this study. C.-Y.H., Y.R., and D.T. assisted with the interpretation of the results. C.-Y.H. and K.-W.H. revised the manuscript for intellectual content.

The datasets generated or analyzed during the current study are not publicly available due to the ethical regulations of the National Taiwan University Hospital but are available from the corresponding author on reasonable request.