Circulating tumour cell counts and ultrasomics signature-based nomogram for preoperative prediction of early recurrence of hepatocellular carcinoma after radical treatment

Abstract

To develop and validate a nomogram incorporating circulating tumour cell counts (CTCs) and the ultrasomics signatures of contrast-enhancement ultrasound (CEUS) for predicting postoperative early recurrence (ER) of HCC after radical treatment.

Methods

Between December 2017 and December 2018, 153 HCC patients (134 males and 19 females; mean age, 56.0 ± 10.2 years; range, 28–78 years) treated with radical therapy were enrolled in our retrospective study and were divided into a training cohort (n = 107) and a validation cohort (n = 46). All patients underwent preoperative CTC tests and CEUS examinations before treatment. The ultrasomics signature was extracted and built from CEUS images. Univariate and multivariate logistic regression analyses were used to identify the significant variables related to ER, which were then combined to build a predictive nomogram. The performance of the nomogram was evaluated by its discrimination, calibration and clinical utility. The predictive model was further evaluated in the internal validation cohort.

Results

HBV DNA, serum AFP level, CTC status, tumour size and ultrasomics score were identified as independent predictors associated with ER (all p < 0.05). Multivariable logistic regression analysis showed that the CTC status (OR = 7.02 [95% CI, 2.07 to 28.38], p = 0.003) and ultrasomics score (OR = 148.65 [95% CI, 25.49 to 1741.72], p < 0.001) were independent risk factors for ER. The nomogram based on ultrasomics score, CTC status, serum AFP level and tumour size exhibited C-indexes of 0.933 (95% CI, 0.878 to 0.988) and 0.910 (95% CI, 0.765 to 1.055) in the training and validation cohorts, respectively, fitting well in calibration curves. Decision curve analysis further confirmed the clinical usefulness of the nomogram.

Conclusion

The nomogram incorporating CTC, ultrasomics features and independent clinical risk factors achieved satisfactory preoperative prediction of ER in HCC patients after radical treatment.

Advances in knowledge

1. CTC status and ultrasomics score were identified as independent predictors associated with ER of HCC after radical treatment. 2. The nomogram constructed by ultrasomics score generated by 17 ultrasomics features, combined with CTCs and independent clinical risk factors such as AFP and tumour size. 3. The nomogram exhibited satisfactory discriminative power, and could be clinically useful in the preoperative prediction of ER after radical treatment in HCC patients.

Introduction

Hepatocellular carcinoma (HCC) is the fourth highest cause of cancer-related mortality, causing a heavy disease burden worldwide. ¹ According to the Barcelona clinical liver cancer (BCLC) staging system, HCC patients in the very early and early stages (BCLC 0 /A) are theoretically ideal candidates for curative treatment, including surgical resection, liver transplantation or local ablation. ^2–4 However, due to the high incidence of early recurrence (ER), the prognosis of HCC is still far from satisfactory. ^5–7 The 2-year recurrence-free survival (RFS) rate for patients with BCLC 0 or A tumours are approximately 30 and 50%, respectively. ^8,9 Early detection and prompt treatment of HCC recurrence can improve prognosis. ^10,11 Therefore, it is necessary to find prognostic biomarkers to monitor postoperative ER in HCC patients.

Recently, the detection of circulating tumour cells (CTCs) before radical hepatectomy has revealed an increased risk of HCC recurrence. ^12–14 CTCs spread from the primary tumour sites or metastases to the peripheral blood supply, and have the characteristics of stem cells combined with invasion capabilities. It is reported that distant metastases caused by CTC invasion are responsible for most recurrence and cancer-related deaths. ^15,16 A CTC ≥ 2 was proved to be a strong factor for predicting ER in HCC after curative resection. ¹⁵

Evidence from the literature also shows the metastatic characteristics of the primary tumour is related to the ER of tumours. ^17,18 Contrast-enhanced ultrasound (CEUS) has been proved to be useful for characterizing liver tumors. ¹⁹ It can reflect the microblood perfusion in tumor with much higher temporal resolution than other imaging modalities. ²⁰ CEUS has been used to monitor the early intrahepatic recurrence of primary HCC after curative treatment with the enhancement patterns of “fast-in and fast-out” or “fast-in and slow-out” and arterial vascularization in the arterial phase. ^21,22 In these studies, most of these imaging features were recognizable to the naked eye. However, due to limited grayscale in visual image, some valuable feature information may be not detected, which restrict the ability to identify useful microscopic image features for disease information. Medical images are subjected to quantitative computational features to reveal disease characteristics which are not visible to the naked eye, known as radiomics. ^23–26 Ultrasomics is a branch of radiomics that uses ultrasound (US) images to extract quantitative features, elucidating the relationship between image features and disease status. ^27,28 A recent study showed that ultrasomics signatures are independent predictors of microvascular invasion in HCC, indicating aggressive behaviour and poor survival outcomes of HCC. ²⁹

We aimed to develop and validate a nomogram incorporating ultrasomics signatures from pre-treatment CEUS images and CTCs together to predict ER in HCC after radical treatment, and we used an independent validation cohort to assess its prediction accuracy.

Methods and materials

Study cohort

The study was approved by the institutional review board of our hospital and was conducted by searching for electronic medical records. Informed consent was waived for this retrospective research. Between December 1, 2017, and December 31, 2018, patients with HCC who underwent radical treatment at our institution were retrospectively included. According to the American Association for the Study of Liver Diseases guidelines, the diagnosis of HCC was based on pathological findings of surgery or biopsy, or clinical diagnosis of dynamic imaging studies and alpha-fetoprotein (AFP) serology. ³⁰ The size and number of HCC tumors were confirmed by performing imaging studies, including computed tomography (CT) or magnetic resonance imaging (MRI). The staging of HCC was assessed using the Barcelona Clinic Liver Cancer (BCLC) staging system. ³¹ The inclusion criteria were as follows ¹ : very early/early-stage tumour, ² Child-Pugh class A or B, and ³ patients were treated with liver transplantation, partial hepatectomy or radiofrequency ablation. The exclusion criteria were as follows ¹ : preoperative antitumour therapy (i.e., repeat liver resection, local ablation and transarterial chemoembolization), ² combination of other malignancies, ³ incomplete response or recurrent HCC close to the resection margin or ablation zone (< 0.5 cm) on radiological imaging, ⁴ lack of CEUS or CTC tests within 7 days before treatment, ⁵ target lesions not completely visible on ultrasound images, and ⁵ clinical-pathologic or follow-up data were not available. Therefore, 153 patients (134 males and 19 females; mean age, 56.0 ± 10.2 years; range, 28–78 years) were identified for final analysis. The study population was randomly divided into a training cohort (n = 107) and a validation cohort (n = 46) at a ratio of 7:3. The training cohort was used to compose models that were evaluated in the validation cohort. Preoperative clinical data were collected for enrolled patients, such as demographic data, serological results and clinical diagnosis. These variables included age, gender, etiology of HCC, maximum tumor size, number of tumors, alanine aminotransferase (ALT), total bilirubin (TBIL), serum albumin (ALB), platelet (PLT) count, Prothrombin time (PT), serum AFP levels, Child-Pugh score and ALBI grade.

CTCs test

CTCs were isolated and counted using negative enrichment and immunofluorescence in situ hybridization technologies (imFISH). ³² In order to deplete the serum, 3.2 ml of peripheral blood was washed with CS1 buffer (Cyttel Biosciences INC., Jiangsu, China), centrifuged for 5 min at 650 g and lysed with CS2 (Cyttel Biosciences INC., Jiangsu, China) to obtain red blood cells. Leukocytes were isolated from the remaining cell pellet in CS1 buffer using immunomagnetic beads conjugated with anti-CD45 monoclonal antibody. CTCs were detected by immunostaining and FISH with antihuman CD45 and a chromosomal centromere-specific probe. Those with CD45-negative cells, centromere of chromosome 8-positive, and 4’,6-diamidino-2-phenylindole-positive samples were considered as CTC-positive. As described in the previous report, the cut-off value for distinguishing CTC-positive from CTC-negative was set at a CTC count of 2.0. ³³

Image acquisition

US examinations were performed using Aplio i500 ultrasound devices (Canon Medical Systems, Tokyo, Japan) equipped with a 375BT convex transducer (frequency, 1.9–6.0 MHz) by one of the two skilled radiologists who had at least 10 years of experience in liver ultrasound. First, the entire liver was scanned with conventional B-mode US. The imaging settings, such as the gain, depth and focus, were optimized for each examination. Then, the transducer was fixed at the largest cross-section of the lesion. After activation of CHI mode, a bolus injection of 2.4 ml of SonoVue (Bracco, Milan, Italy) was administered, followed by flushing with 5 ml of saline. The target lesion was continuously observed for at least 90 s without any change in machine settings. After 90 s, the lesion was intermittently scanned and 3–5 min digital cine clips were stored on the hard disk. The arterial, portal venous and late phases were defined as 10–30 s, 20–120 s and 121–300 s after injection, respectively. ³⁴

Ultrasomics feature extraction and selection

For each lesion, a region of interest (ROI) around the tumour outline was manually delineated on the largest cross-section of US images using the ITK-SNAP software (open-source software; http://www.itksnap.org) (Figure 1). The ultrasomics features were extracted using in-house designed Ultrasomics-Platform software (Ultrasomics Artificial Intelligence X-lab, Guangzhou, China). A total of 1044 features were automatically extracted from a single image and were divided into five categories: grey-level histogram feature, transformed matrix texture, wavelet transformed feature, shearlet transformed feature and gabor filter transformed feature. Detailed information on the features is provided in Supplementary Material S1. All ultrasomics feature values were normalized to be within similar dynamic ranges using the mean and variance of the feature values. Finally, 4176 features were extracted from the baseline US, arterial phase, portal phase and delay phase of CEUS images of each patient. A test–retest procedure was performed on a cohort of 30 randomly chosen patients. Two radiologists independently performed the segmentation to evaluate the reproducibility of the extracted features. The features with intraclass correlation coefficients lower than 0.80 were excluded.

Figure 1.

Illustration of original image and segmentation. Annotations of the region of interest (ROI) generated by the radiologists around the tumour outline are delineated in red.

To reduce the ultrasomics features dimensions and identify the features that were highly effective for predicting ER in HCC patients, the least absolute shrinkage and selection operator (LASSO) regression algorithm was performed (Figure 2). Adjusted by λ, the LASSO method can shrink all coefficients to zero and set the coefficients to zero of irrelevant features. Finally, 17 ultrasomics features (6 features from 2D plain images, 7 features from arterial phase images and 4 features from portal venous phase images) were used to generate the ultrasomics score formula. Detailed information on the selected features is shown in the supporting information (Supplementary Material S2, Supplementary Table S1). The ultrasomics score was calculated using the formula described in Supplementary Material S3 (Supplementary Table S2).

Figure 2.

Ultrasomics feature selection using the least absolute shrinkage and selection operator (LASSO) regression model in the training cohort.

Follow-up and recurrence

Patients were followed up at 1 month after curative treatment and every 3 months thereafter, based on AFP and imaging examinations. Contrast-enhanced computed tomography (CECT) or gadoxetic acid–enhanced MRI was performed every 3 months. ER was defined as new intrahepatic lesions and/or extrahepatic metastasis with typical imaging features or/and with histopathological confirmation, within 2 years after curative treatment. ³⁵ The endpoint was evidence of recurrence. Patients with no recurrence were followed up until December 31, 2020.

Development and validation of ER-predicting nomograms

Univariate and multivariate logistic regression analyses were performed in the training cohort to identify the independent risk factors for ER. In multivariate logistic regression analysis, variables with p < 0.05 in the univariate analysis were included. Additionally, P values of less than 0.05 were considered significant in the multivariate analysis. Finally, a clinical-radiological nomogram incorporating the ultrasomics score, CTCs and independent clinical risk factors, which could visually and individually identify the probability of ER in the training cohort, was constructed based on univariate and multivariate logistic regression.

The discriminative performance of the nomogram was quantified in the independent validation group using the Harrell’s concordance index (C-index), which ranges from 0 to 1. The calibration curves were plotted using observed probabilities and the nomogram-predicted probabilities. ^36,37 A decision curve analysis (DCA) was performed in the validation cohort to quantify the net benefits at different threshold probabilities. ³⁸

Statistical analysis

Categorical variables were compared using the chi-square test or Fisher’s exact test, while continuous variables were compared using Student’s t test for variables with a normal distribution or the Mann-Whitney U-test for variables with an abnormal or unknown distribution. In the training cohort, univariable and multivariable logistic regression analyses were performed to identify the independent risk variables for ER. To construct radiomics nomogram, variables with p < 0.05 in the univariate analysis were included in the multivariable analysis. Multivariable analysis was conducted using the stepwise backward elimination method. The odds ratios (ORs) were calculated using 95% confidence intervals (CIs) for each risk factor.

Statistical analysis was performed with R software v4.0.2 (R Foundation for Statistical Computing, Vienna, Austria). The reported statistical significance levels were all two-sided, and p < 0.05 was considered statistically significant. The “glmnet” package was used for LASSO logistic regression. The “glm” function was used for the univariate and multivariate logistic regression analyses. The “Hmisc” package was used to plot the nomogram and quantified the C-index. The “calibration_curves” package was used for the calibration curves. The “decision_curve” package was used to perform DCA.

Results

Clinical characteristics

Patient characteristics in the training and validation cohorts are summarized in Table 1. In total, 153 patients were included in the final study group and divided into the training cohort (n = 107) and validation cohort (n = 46). Positive CTCs were accounted for 61.4% (94/153) of patients. There was no significant difference between the two cohorts in the presence of CTCs (p = 0.783). In the entire cohort, the median follow-up time was 18.0 months (95% CI, 16.0–19.0 months). Among the included patients, ER was identified in 74 (48.4%) patients (intrahepatic recurrence: n = 61; extrahepatic recurrence: n = 13). The median follow-up time of no ER group and ER group were 21.5 months (95% CI, 20.0–27.0 months) and 8.0 months (95% CI, 6.0–9.0 months), respectively. There was no difference in the ER rate between the training cohort (54/107, 50.5%) and the validation cohort (20/46, 43.5%, p = 0.752). In addition, there were no significant differences between the two cohorts in other characteristics.

Table 1.

Basic characteristic of patients in the training and validation sets

Characteristic	Training Cohort (n = 107)	Validation Cohort (n = 46)	P value
Patient demographics
Age (year) a	56.9 ± 9.3	54.2 ± 11.6	0.438
Gender			0.674
Female	12 (11.2)	7 (15.2)
Male	95 (88.8)	39 (84.8)
Aetiology of HCC			0.669
HBV	86 (80.4)	35 (76.1)
HCV	6 (5.6)	2 (4.3)
Other	15 (14.0)	9 (19.6)
Laboratory parameters
HBV-DNA(IU/ml)			0.203
≤100	66 (61.7)	34 (73.9)
>100	41 (38.3)	12 (26.1)
ALT (IU/L)	30.5 [20.0–50.5]	33.0 [20.0–53.5]	0.349
TBIL (μmol/L) b	15.3 [12.7–22.7]	14.5 [11.8–19.9]	0.346
ALB(g/dL) a	36.8 ± 4.8	36.5 ± 4.9	0.728
PLT (×10^9  L−1) b	143 [95.3–172.0]	145.0 [74.0–216.0]	0.717
PT (sec) b	12.3 [11.0–13.1]	12.1 [11.2–13.9]	0.582
Serum AFP level (ng/mL)			0.611
<20	59 (55.1)	23 (50.0)
20–400	31 (29.0)	17 (37.0)
>400	17 (15.9)	6 (13.0)
Child-Pugh score			0.202
A	94 (87.8)	36 (78.2)
B	13 (12.1)	10 (21.7)
ALBI grade			0.630
1	8 (7.5)	4 (8.7)
2	80 (74.8)	31 (67.4)
3	19 (17.7)	11 (23.9)
US features
Tumour number			0.142
1	101 (94.4)	39 (84.8)
2	3 (2.8)	3 (6.5)
3	3 (2.8)	4 (8.7)
Tumour size (cm)b	3.3 [1.9–4.9]	3.2 [2.1–5.7]	0.689
Treatment			0.252
Resection	36 (33.6)	11 (23.9)
Transplantation	17 (15.9)	12 (26.1)
Ablation	54 (50.5)	23 (50.0)
CTC status			0.783
＜2	40 (37.4)	19 (41.3)
≥2	67 (62.6)	27 (58.7)
Early recurrence	54 (50.5)	20 (43.5)	0.752

Note---Data are numbers of patients, with percentage in parentheses unless indicated. HBsAg = hepatitis B surface antigen status, HBV-DNA = hepatitis B virus DNA load, ALT = alanine aminotransferase, TBIL = total bilirubin, ALB = albumin, PLT = platelet count, PT = prothrombin time, AFP = alpha-fetoprotein, ALBI = albumin-bilirubin

^aData are mean ± standard deviation

^bData are medians, with interquartile ranges in parentheses

Evaluation of the ultrasomics signature, CTCs and clinical risk factors

Table 2 shows the results of the univariate and multivariate logistic regression analyses for ER in the training cohort. HBV DNA, serum AFP level, CTC status, tumour size, and ultrasomics score were identified as independent predictors associated with ER (all p < 0.05). In the multivariable logistic regression analysis, the CTC status (OR, 7.02, 95% CI, 2.07–28.38, p = 0.003) and ultrasomics score (OR, 148.65, 95% CI, 25.49–1741.72, p < 0.001) significantly predicted ER.

Table 2.

Univariate and multivariate regression analyses between early recurrence and non-recurrence groups in the training cohort.

Parameter	Univariate analysis		Multivariate analysis
	Hazard Ratio	P value	Hazard Ratio	P value
Age	0.98 (0.94, 1.01)	0.213
Gender	1.44 (0.43, 4.78)	0.556
HBsAg	0.96 (0.37, 2.52)	0.934
HBV-DNA	2.75 (1.22, 6.22)	0.015a	1.24 (0.25, 6.20)	0.794
PLT	1.00 (0.99, 1.01)	0.050
ALT	1.00 (0.99, 1.01)	0.646
ALB	1.02 (0.94, 1.10)	0.676
PT	1.01 (0.90, 1.15)	0.823
Serum AFP level
<20
20–400	2.70 (1.10, 6.64)	0.031a	1.36 (0.26, 7.19)	0.719
>400	7.59 (2.15, 26.84)	0.002a	3.96 (0.44, 35.71)	0.220
Child-Pugh score	0.39 (0.11, 1.36)	0.126
ALBI grade
1
2	1.70 (0.67, 4.29)	0.265
3	0.32 (0.03, 3.03)	0.318
CTC status	5.16 (2.23, 11.92)	<0.001a	7.02 (2.07, 28.38)	0.003a
Tumour size	1.19 (1.04, 1.37)	0.014a	1.01 (0.79, 1.29)	0.924
Tumour number
1
2		0.972
3		0.974
Ultrasomics score	412.17 (43.91, 3868.60)	<0.001a	148.65 (25.49, 1741.72)	<0.001a
Treatment
Resection
Transplantation	0.89 (0.39, 2.02)	0.780
Ablation	0.46 (0.15, 1.41)	0.176

Note.—Data in parentheses are the 95% confidence intervals. Significant variables with p < 0.10 in the univariate analysis were included in the multivariate logistic regression analysis.

^a P value < 0.05

Nomogram construction and predictive performance

Based on univariate and multivariate logistic regression, a predictive nomogram that incorporated AFP, tumour size, CTCs and ultrasomics scores was constructed (Figure 3). Figure 4a illustrates the calibration curve of the proposed nomogram based on the training cohort. In the training cohort, the C index values of the CTCs, ultrasomics score, and nomogram were 0.604 (95% CI, 0.489–0.719), 0.876 (95% CI, 0.764–0.903) and 0.933 (95% CI, 0.878–0.988), respectively. Moreover, a favourable calibration (Figure 4b) was confirmed in the validation cohort. The nomogram demonstrated a significantly higher discriminative power (AUC, 0.910, 95% CI, 0.765–1.000) than the CTCs (AUC, 0.583, 95% CI, 0.411–0.756, p < 0.001) and the ultrasomics score (AUC, 0.839, 95% CI, 0.748–0.956, p = 0.012) in the validation cohort. The clinical utility of the nomogram, ultrasomics score and CTCs was evaluated using DCA (Figure 5). The DCA curves showed that if the threshold probability was>10%, using the nomogram and ultrasomics score added more net benefit for patients than using CTCs. When the threshold probability was >40%, the nomogram increased the benefit to patients over that using CTCs and the ultrasomics score.

Figure 3.

Nomogram for predicting early recurrence probabilities in HCC patients after radical treatment.

Figure 4.

Calibration curves of the nomogram in the training (a) and validation (b) cohorts; the X-axis is the nomogram-predicted probability of early recurrence. The Y-axis is the actual early recurrence, and the diagonal-dashed line indicates the ideal prediction by a perfect model.

Figure 5.

Decision curve analysis (DCA) derived from the validation cohort. The Y-axis measures the net benefit. The net benefit is determined by calculating the difference between the expected benefit and the expected harm associated with each proposed model [net benefit = true positive rate − (false positive rate×weighting factor), weighting factor = threshold probability/ (1-threshold probability)]. The grey line represents the assumption that all patients had early recurrence. If the threshold probability was >10%, using the nomogram (red curve) and ultrasomics score (green curve) added more net benefit for patients than using CTCs (blue curve). When the threshold probability was >40%, the nomogram (red curve) increased the benefit to patients over that using CTCs (blue curve) and ultrasomics score (green curve).

Discussion

The recurrence of HCC after radical treatment has become one of the main factors affecting prognosis and survival. ³⁹ It is generally believed that ERs originate from intrahepatic metastases of primary tumours and are mainly related to tumour biological factors, such as tumour size, number of lesions, vascular invasion and differentiation. ^35,40 Studies have shown that peripheral CTC detection can describe the migration of CTCs from the primary tumour to the peripheral blood, which reveals the mechanisms of metastasis and facilitates the prediction of postoperative recurrence or metastasis in HCC. In addition to the detection of CTCs, the development of imaging techniques has offered notable advantages over existing prognostic sources for HCC. ⁴¹ Herein, we demonstrated that both CTC and ultrasomics features were associated with ER in HCC. Furthermore, we constructed a novel nomogram incorporating CTC and ultrasomics features derived from CEUS for preoperatively predicting ER of HCC. The nomogram exhibited good prediction performance (C index, 0.910), calibration and clinical application in an independent validation cohort. Therefore, it can be used as a noninvasive and effective tool to preoperatively identify patients at higher risk of ER after radical treatment.

CTCs are considered to be responsible for tumour recurrence and metastasis. ⁴² They have vascular invasion ability and tend to circulate even in early-stage tumours. They have been studied in the context of screening, diagnosis, and surveillance. In our study, CTCs ≥ 2 were an independent risk factor for ER, which is consistent with other reports. Sun et al reported that CTCs retaining stem cell-like characteristics are “high-quality seeds” for metastasis and that CTCs ≥ 2 is an independent predictor for postoperative ER and prognosis of HCC. ⁴³ Felden et al and Zhou et al reported that EpCAM-positive CTCs prior to curative-intended liver resection disclose an elevated risk of HCC recurrence which provides a novel prognostic predictor for HCC patients. ^13,14 As a form of liquid biopsy, CTCs have great potential to facilitate the implementation of precision medicine in patients with HCC. We expect the ongoing clinical trial (NCT02973204) to further confirm CTCs as a clinical support tool in HCC.

Most ERs are likely to result from intrahepatic metastasis disseminated from the primary tumour. Thus, the aggressive characteristics of the primary tumour are one of the determinants for ER. Imaging, characterizing the physiologic and molecular features of tumours, plays an imperative role in the field of oncology. ⁴⁴ Radiomics, focused on image analysis by extracting hundreds of quantitative features with a computer algorithm, could enable quantification of tumour heterogeneity by representing the spatial arrangement of imaging voxels with signal intensity variations. In our study, we established the ultrasomics score by extracting 17 ultrasomics features that were closely associated with ER of HCC from US and CEUS images, which included three grey-level histogram features, three transformed matrix textures, and 11 shearlet transformed features. It is worth noting that shearlet transformed features achieved the highest weights in our signature. These features are more effective at identifying sparse anisotropic edges, with an existing automatic wavelet-based registration algorithm. ⁴⁵ Accordingly, ultrasomics features can provide more detailed information on tumour biology as well as the tumour microenvironment that is complementary to visual features.

In our study, the ultrasomics score demonstrated satisfactory discriminative power in both the training and validation cohorts (AUC = 0.876 and 0.839, respectively), which was indicated to be an independent predictor for ER (p < 0.001). We further integrated the CTCs into the ultrasomics score to present a nomogram as a tool for individualized risk estimation and obtained higher discriminative power in both the training and validation cohorts (AUC = 0.933 and 0.910, respectively).

This nomogram is clinically relevant because it can identify a small number of patients at high risk of ER that may be easy to manage. National Comprehensive Cancer Network guidelines recommend the surveillance after curative therapy for HCC imaging every 3–6 months for 2 years, then every 6–12 months thereafter. ⁴⁶ For patients with high risk of ER after radical treatment for HCC, close postoperative surveillance is necessary. ⁴⁷ Although it may not be appropriate to exclude patients with a high risk of ER from radical surgery, first adjuvant therapy should be given priority for these patients. ¹¹

Our study has several limitations. First, it was a retrospective analysis and performed in a single institution that suffers from inherent biases and limits its generalizability. Second, although all CEUS images in this study were acquired in a uniform US scanner with standardized sequences for imaging acquisition to reduce bias and variance in our results, operator dependency may still influence the ultrasomics features, further external validation from other institutions with different scanners is warranted to check for generalizability. What’s more, when acquiring the CEUS images, the transducer was fixed at the largest cross-section of the lesion. Thus, the ultrasomics features were only extracted on the images of the largest cross-section instead of the 3D volumetric images. Moreover, only CEUS images were used to extract ultrasomics features instead of other imaging modalities, such as CT or MRI, which may limit the clinical utility of this nomogram. Radiomics models generated from multi-imaging modalities may be constructed in further studies. Third, although negative enrichment does not depend on any physical properties of tumour cells and ImFISH further improves the sensitivity and signal intensity, due to the heterogeneity of HCC, current CTC detection methods still have a relatively low number of detected CTCs. Therefore, continuous improvements in methods related to cell separation and identification are needed. Fourth, the different treatment modalities were not analysed separately. Treatment modalities weren’t identified as independent predictors associated with ER in the univariate logistic regression. Further investigation with large sample size may performed to compare the subgroup between different treatment.

Conclusions

In conclusion, this study demonstrated that the nomogram incorporating CTCs, ultrasomics features and clinical factors demonstrated satisfactory discriminative power, and could be clinically useful in the preoperative prediction of ER after radical treatment in HCC patients.