The value of virtual non-contrast images derived from dual-energy spectral CT in the short-term efficacy assessment of hepatocellular carcinoma after TACE.

1
Introduction
Hepatocellular carcinoma (HCC) is the most common malignant liver tumor. Patients often present at an intermediate or advanced stage when diagnosed due to atypical incipient symptoms and a lack of routine surveillance, thereby losing the opportunity for curative treatment [1], [2]. Transarterial chemoembolization (TACE) is the preferred treatment for intermediate-stage HCC, particularly in patients who are not candidates for liver transplantation [3]. The determination of early therapeutic efficacy in HCC after TACE facilitates the formulation of subsequent treatment strategies. Thus, imaging evaluation of the residual viable tumor (RVT) and lipiodol deposition is crucial after TACE.
Contrast-enhanced MRI outperforms contrast-enhanced CT in detecting RVT in the early postoperative period in HCC patients after TACE, attributable to its superior soft-tissue resolution. However, it is less sensitive in displaying lipiodol deposition [4], which is a factor closely related to tumor necrosis and patient prognosis [5], [6]. In comparison, contrast-enhanced CT simultaneously displays lipiodol deposition and evaluates RVT after TACE, making it the preferred imaging tool for postoperative follow-up. However, the polychromatic energy images of a conventional contrast-enhanced CT may give rise to ray beam-hardening artifacts combined with a shielding effect of lipiodol on RVT and adjacent structures [7], thus weakening the ability of conventional CT to precisely evaluate TACE efficacy.
Dual-energy spectral CT (DEsCT) allows for the identification of iodine-containing voxels and then measures and subtracts their iodine component to generate virtual non-contrast (VNC) images that closely resemble true non-contrast (TNC) images. Previous research has reported that VNC images show good consistency with TNC images [8]. Furthermore, they enable reliable identification of renal lesions and pancreatic diseases, while also proving clinically valuable in the differential diagnosis of adrenal diseases [9], [10], [11]. These findings suggest that VNC images can serve as an alternative to TNC images for diagnostic purposes while reducing radiation exposure. Additionally, studies have reported that lipiodol within HCC after TACE, which appears hyperattenuating on TNC images, can be suppressed on VNC images [12]. However, the application value of such lipiodol-removing capability of VNC images in evaluating post-TACE therapeutic efficacy for HCC remains insufficiently explored.
Therefore, this study aims to evaluate the ability of VNC images to reduce lipiodol and associated artifacts and determine their diagnostic efficacy for different components in the operative area and normal hepatic parenchyma in HCC after TACE, thereby exploring their value in the short-term clinical follow-up of HCC patients after TACE.

2
Methods
2.1
General information
This retrospective study, approved by the Institutional Review Board (approval number: KY-2021–052–03, KY-2024–5–89–1, KYLL-2024B-039), included 94 HCC patients with TACE who underwent DEsCT scans within 4–6 weeks after treatment from November 2020 to May 2025. Due to its retrospective nature, informed consent was waived. Patients were retrospectively selected based on clinical, imaging diagnoses from four hospitals.
The inclusion criteria were as follows: 1. Patients diagnosed with HCC through clinical, imaging examinations; 2. Patients underwent TACE and had DEsCT scans 4–6 weeks after treatment; 3. Presence of both lipiodol and RVT within the HCC lesions; 4. The size and morphology of the RVT and lipiodol allowed for the delineation of regions of interest (ROI); 5. Patients had not undergone surgical resection, liver transplantation, ablation, radiotherapy, or systemic therapies before or following TACE or during follow-up period.
The exclusion criteria were as follows: 1. No RVT was observed after TACE; 2. The longest diameter of the RVT was less than 5 mm, affecting accurate data measurement; 3. Poor lipiodol deposition after TACE, affecting accurate ROI delineation; 4. Severe image artifacts.
After screening, 66 patients were enrolled. The enrollment process is shown in Fig. 1.

2.2
TACE treatment
TACE was performed in accordance with standard clinical protocols [13]. Specifically, an emulsion of chemotherapeutic agents and lipiodol was injected into the tumor-feeding artery via femoral artery puncture. The specific treatment regimen was individualized based on the tumor size, number, and arterial blood flow.

2.3
CT examination
Upper abdominal TNC and contrast-enhanced CT scans were performed using either a Revolution CT or a Discovery CT 750HD scanner (GE HealthCare, Wisconsin, USA). Patients fasted for 6 h before the examination. The TNC scanning was performed in the conventional scanning mode, while the arterial phase (AP) and portal venous phase (PVP) scans were conducted using the gemstone spectral imaging (GSI) mode. Smart triggering technology was utilized for the contrast-enhanced CT scan, with the ROI positioned on the abdominal aorta. The scan was triggered 7 s after the enhancement in the ROI reached the threshold of 100 HU, and the PVP scans began 30 s after the AP scans were completed. The contrast agent used was Iohexol (350 mg I/ml), administered as a 70 ml bolus injection through the antecubital vein at a flow rate of 5 ml/s, followed by a 50 ml saline flush.
The scanning parameters were as follows: Both the TNC and contrast-enhanced scans employed helical scanning with a tube rotation speed of 0.5 s/r, a collimator width of 80 mm, and a pitch of 0.992:1. For the TNC scans, the tube voltage was set at 120 kVp with automatic tube current modulation. For the contrast-enhanced scans, the GSI mode with fast tube voltage switching between 80 and 140 kVp during gantry rotation was applied, and the tube current was fixed at 405 mA.
The TNC and contrast-enhanced images were reconstructed to a slice thickness and interval of 1.25 mm and transferred to an AW 4.7 workstation (GE HealthCare, Wisconsin, USA). The material suppressed iodine (MSI) module on the workstation was used to convert the AP and PVP images into corresponding VNC images (VNCAP and VNCPVP, respectively). All image analyses and data measurements were performed on the 1.25 mm images. For patients with multifocal HCC, the largest lesion was selected for image analysis and data measurement.

2.4
Image analysis and data measurement
2.4.1
Subjective scoring of beam-hardening artifacts surrounding lipiodol
Two radiologists with 3 and 5 years of experience in abdominal imaging diagnosis subjectively scored the degree of beam-hardening artifacts surrounding the lipiodol on TNC, VNCAP, and VNCPVP images using a 5-point Likert scale: 1 = no artifacts, 2 = minor artifacts, 3 = moderate artifacts, 4 = severe artifacts, and 5 = massive artifacts [14]. The result was determined by consensus through discussion between two radiologists for cases with disagreement.

2.4.2
Assessment of the lipiodol removal degree
The same two radiologists subjectively evaluated the lipiodol removal degree within the HCC lesions after TACE on VNCAP and VNCPVP images by comparing with TNC images and classified it into 4 grades:
Grade 1: VNC images removed ≤ 25 % of the lipiodol; Grade 2: VNC images removed 26 %-50 % of the lipiodol; Grade 3: VNC images removed 51 %-75 % of the lipiodol; and Grade 4: VNC images removed > 75 % of the lipiodol.
The two radiologists were blinded to the patients' clinical information and assessed the TNC, VNCAP, and VNCPVP images comprehensively using axial, coronal, and sagittal views reconstructed in three dimensions. When there were inconsistencies, decisions were made in consensus.

2.4.3
ROI delineation and data measurement
A radiologist with 6 years of experience in abdominal imaging diagnosis delineated the ROIs for RVT, adjacent normal hepatic parenchyma (ANHP), and lipiodol removal area (LRA) according to the following criteria.
This study utilized 70-keV upper abdominal contrast-enhanced DEsCT images, which demonstrated equivalent performance to 120-kVp polychromatic CT images [15], for RVT assessment. RVT was identified as areas showing hyperenhancement in AP and washout in PVP or delayed phase on contrast-enhanced CT images within the postoperative region [16]. For RVT indeterminate on contrast-enhanced CT, definitive diagnosis was established through extracellular contrast-enhanced MRI conducted within a two-week interval following DEsCT examination. The imaging interpretations were validated by a senior abdominal radiologist with > 20 years of experience in abdominal imaging diagnosis. ANHP was defined as hepatic parenchyma > 1 cm from the tumor margin to avoid the influence of microvascular infiltration from RVT [17]. LRA was defined as regions demonstrating definitive lipiodol deposition on TNC images that became undetectable at corresponding locations on both VNCAP and VNCPVP images.
ROIs for RVT and ANHP were initially delineated on the 70-keV DEsCT images in AP. Since TNC images can optimally depict lipiodol deposition morphology before its subtraction on VNC images, ROIs for LRA were more accurately defined when first delineated on TNC images. Following initial delineation, all ROIs were pasted across TNC, VNCAP, and VNCPVP images using the workstation's copy-and-paste function to ensure spatial consistency for CT attenuation value quantification. When respiratory motion-induced misregistration occurred, corresponding ROIs were manually adjusted. To ensure measurement reliability, all CT attenuation value measurements were performed three times independently, and the average values were used as the results for statistical analysis.
All ROIs were geometrically standardized as either circular or elliptical. For RVT and LRA, ROI sizes were approximately 50–150 mm², while smaller LRA volumes required customized manual delineation along their edges to ensure precise lesion coverage. ROIs for ANHP were approximately 200–250 mm². ROIs for RVT and ANHP were placed avoiding lipiodol, blood vessels, artifacts, and inhomogeneous liver parenchyma.
CT attenuation values of RVT, ANHP, and LRA were compared among TNC, VNCAP, and VNCPVP images. The diagnostic performance of CT attenuation values for the three areas was evaluated and compared between VNCAP and VNCPVP images.

2.5
Statistical analysis
Statistical analyses were performed using SPSS (Version 27.0, IBM Corporation) and MedCalc (Version 22.001).
The consistency of the scores on beam-hardening artifacts on TNC, VNCAP, and VNCPVP images, as well as lipiodol removal degree on VNCAP and VNCPVP images between the two observers was compared using a Kappa test. The kappa values were interpreted as follows: < 0.2, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.0, perfect agreement. Comparisons of beam-hardening artifact scores between TNC, VNCAP, and VNCPVP images, as well as comparisons of lipiodol removal degree between VNCAP and VNCPVP images were performed using the chi-square test, with P < 0.05 indicating statistical significance.
CT attenuation values of RVT, ANHP, and LRA were tested for normality. Normally distributed continuous data were expressed as mean ± standard deviation, and one-way analysis of variance (ANOVA) was used to compare differences among the three groups, with posthoc analysis using the LSD test for pairwise comparisons. Non-normally distributed continuous data were expressed as median (interquartile range), and the Kruskal-Wallis test was used to compare differences among the three groups, with pairwise comparisons using the Mann-Whitney U test. P values for intergroup comparisons were adjusted using the Bonferroni method, with P < 0.05 considered statistically significant.
Receiver operating characteristic (ROC) curves were constructed to assess the diagnostic performance of the CT attenuation values on VNCAP and VNCPVP images for identifying RVT, ANHP, and LRA. The DeLong test was used to compare the diagnostic performance of CT attenuation values between VNCAP and VNCPVP images for differentiating these areas.

3
Results
3.1
General information
After screening, 66 HCC patients (56 males, 10 females; age range 24–84 years, median 61 years) with TACE were enrolled. Among them, 27 patients underwent contrast-enhanced MRI within 2 weeks after DEsCT examination. A total of 66 lesions were evaluated in this study. Detailed data of patients and lesions are listed in Table 1.

3.2
Comparison of subjective scoring of beam-hardening artifacts
The Kappa test showed high interobserver consistency (k = 0.90, 0.84, and 0.87) for subjective scoring of the beam-hardening artifacts on TNC, VNCAP, and VNCPVP images, respectively.
The beam-hardening artifact scores surrounding the lipiodol on TNC, VNCAP, and VNCPVP images are shown in Table 2,
Fig. 2. Since the number of lesions with beam-hardening artifact scores of 4 and 5 was zero, they were merged with lesions scored 3 for analysis. There were statistically significant differences among the three types of images (P < 0.001). Further pairwise comparisons revealed that the differences in beam-hardening artifact scores between TNC and VNCAP images and between TNC and VNCPVP images were statistically significant (P < 0.001), while no significant difference was found between VNCAP and VNCPVP images (P > 0.05).
On TNC images, beam-hardening artifacts were present surrounding lipiodol in 34 lesions, of which 28/34 (82.4 %) and 26/34 (76.5 %) were significantly reduced or eliminated on VNCAP and VNCPVP images, respectively, with subjective scores less than 3 (Fig. 3). Only 6 and 8 lesions showed persistent beam-hardening artifacts surrounding lipiodol on VNCAP and VNCPVP images, respectively.

3.3
The lipiodol removal degree on VNCAP and VNCPVP images
The two radiologists showed good interobserver consistency (k = 0.82 and 0.86) for evaluating the lipiodol removal degree on VNCAP and VNCPVP images, respectively. The lipiodol removal degree in all enrolled HCC lesions by both VNC images was ≥ Grade 3, with 7 lesions at Grade 3 and 59 lesions at Grade 4. Complete lipiodol removal was observed in 37.9 % (25/66) of HCC lesions.

3.4
Comparison of CT attenuation values of RVT, ANHP, and LRA in HCC after TACE
The CT attenuation values of RVT, ANHP, and LRA on TNC, VNCAP and VNCPVP images are shown in Table 3. Statistically significant differences in CT attenuation values were observed among the three areas within the same image (P < 0.001).
Pairwise comparisons revealed significant differences in CT attenuation values between LRA and both ANHP and RVT on TNC images (P < 0.001), but no significant difference was observed between RVT and ANHP. Notably, all three areas demonstrated statistically significant differences in CT attenuation values on both VNCAP and VNCPVP images (P < 0.001), with LRA having the lowest CT attenuation values, followed by RVT which remained significantly lower than ANHP (P < 0.001) (Fig. 4,
Table 4).

3.5
Diagnostic performance of CT attenuation values on VNCAP and VNCPVP images
The ROC curves showed that CT attenuation values on VNCAP and VNCPVP images had high efficacy in differential diagnosis of RVT from LRA, RVT from ANHP, and ANHP from LRA, with area under the curve (AUC) values of 0.80, 0.77, and 0.97, respectively using VNCAP and 0.76, 0.84, and 0.98, respectively using VNCPVP images. The DeLong test demonstrated that VNCAP images outperformed VNCPVP images in differentiating RVT from LRA (P < 0.05). Conversely, VNCPVP images showed superior diagnostic performance compared to VNCAP images when differentiating RVT from ANHP (P < 0.05). No statistically significant difference was observed between the two images in differentiating ANHP from LRA (P > 0.05) (Fig. 5,
Table 5).

4
Discussion
Although TACE remains a cornerstone treatment for specific HCC, suboptimal therapeutic responses occur in some patients. Moreover, repeated procedures potentially exacerbate hepatic functional decline [18]. Therefore, an accurate assessment of TACE efficacy is essential for determining the proper TACE candidates and patients’ clinical outcomes. Although contrast-enhanced CT is cost-effective for post-TACE surveillance, lipiodol and associated beam-hardening artifacts reduce its accuracy in detecting RVT and assessing TACE-induced necrosis. Our study demonstrated that VNC images in DEsCT could effectively remove lipiodol and reduce associated beam-hardening artifacts, thereby facilitating the identification of RVT and the assessment of TACE-induced tumor necrosis. Thus, VNC images can synergize with TNC and contrast-enhanced CT images and serve as an effective supplementary tool to enable more reliable therapeutic response evaluation following TACE.
Our study indicated that VNC images could significantly reduce beam-hardening artifacts surrounding lipiodol. Beam-hardening artifacts occur when low-energy photons of a polychromatic x-ray beam are absorbed more by lipiodol [19], consequently obscuring RVT in HCC after TACE on conventional CT. VNC images are reconstructed using a post-processing algorithm that subtracts iodine maps from DEsCT-enhanced images [20]. The material decomposition algorithm for iodine maps is associated with fewer beam-hardening artifacts [21]. Thus, VNC images show potential for beam-hardening artifact reduction. Our findings were similar to the in vitro experimental results reported by Chen et al. [22]. However, some beam-hardening artifacts surrounding the lipiodol persisted on VNC images in our study, possibly due to the imperfect calibrations or incomplete filtration of low-energy photons in DEsCT [23], [24].
In our study, VNC images exhibited effective lipiodol removal capability in HCC after TACE, with no difference in the degree of lipiodol removal between the two VNC images. Iodine confers the characteristic hyperdensity of lipiodol on CT images [25]. Thus, the capacity of VNC images to precisely identify and subtract iodine in lipiodol effectively removes lipiodol [12]. However, partial lipiodol remained persistent on VNC images in this study, likely due to local lipiodol concentration exceeding VNC images’ iodine suppression threshold [24]. Although VNC images could not completely suppress lipiodol interference in all cases, lipiodol-retaining areas might represent sufficient local lipiodol deposition, indicating a favorable treatment response. Notably, while lipiodol deposition serves as an imaging biomarker for treatment response, its hyperdensity interferes with the identification of RVT, the pivotal determinant for evaluating TACE efficacy. Our findings indicated that LRA presented as hypodense areas with significantly lower CT attenuation values than RVT on VNC images. Lipiodol deposition conventionally indicates tumor necrosis in HCC after TACE [26]. Therefore, LRA might correspond to TACE-induced hypodense necrotic tissue. Our results demonstrated that VNC images could effectively reduce lipiodol interference in RVT detection while restoring the density contrast between necrosis and RVT. This capability enables the clear differentiation between RVT and necrotic tumor tissue with precise boundary delineation, thereby improving the accuracy of TACE efficacy assessment.
In HCC patients following TACE, the density contrast between RVT and ANHP may diminish or even disappear on TNC images due to lipiodol, beam-hardening artifacts, and the cirrhotic background [27]. In our study, we observed similar CT attenuation values between RVT and ANHP, along with blurred boundaries, on TNC images. Although contrast-enhanced CT can differentiate RVT by highlighting its enhancement characteristics, lipiodol interference persists. In our study, VNC images overcame this limitation by reducing lipiodol and beam-hardening artifact interference, revealing RVT as hypodense to ANHP through the restoration of their intrinsic CT attenuation value differences. Thus, VNC images can serve as a viable tool to assist in RVT observation independent of enhancement characteristics.
ROC curve analysis substantiated the enhanced diagnostic efficacy of VNCAP images for differentiating RVT from LRA, while VNCPVP images showed better efficacy in differentiating RVT from ANHP. This difference stems from the significant impact of local iodine contrast media distribution patterns on VNC image-derived CT attenuation values, with iodine-rich regions exhibiting higher CT attenuation values than iodine-deficient areas [28], [29]. Specifically, RVT exhibits high AP iodine contrast media uptake. As a result, VNCAP images exhibit more obvious CT attenuation value differences between RVT and LRA. Normal hepatic parenchyma maintains enhancement during the PVP, while RVT exhibits iodine contrast media washout, resulting in more pronounced CT attenuation value differences between the two on VNCPVP images. Although these phenomena might stem from technical limitations of VNC reconstruction algorithms, our results demonstrated that the diagnostic efficacy for RVT can be effectively optimized through the rational application of VNC images from different phases under current technical conditions.
Our study demonstrated comparable efficacy between VNCAP and VNCPVP in differentiating ANHP from LRA, with LRA showing more distinct hypodensity than ANHP, which was similar to the results of Zhang et al. [30]. Wu et al. [26] demonstrated that HCC patients developing liquefactive necrosis after the first TACE procedure had a worse prognosis compared to those with coagulative necrosis. VNC images’ visualization of LRA indicated their ability to directly observe the density characteristics of TACE-induced necrosis. Furthermore, the extent of TACE-induced necrosis represents another critical dimension in treatment response assessment. TNC images enable indirect assessment of TACE-induced necrosis areas by evaluating lipiodol distribution. However, in cases of lipiodol deposition defects, lipiodol or artifacts surrounding the lipiodol-deficient areas may lead to underestimation of defect extent. This may diminish the accuracy of lipiodol distribution assessment using TNC images. In our study, LRA presented as a hypodense region with clear boundaries relative to ANHP on VNC images. This facilitates accurate determination of the extent of TACE-induced necrotic tissue, thereby enabling a more comprehensive evaluation of TACE efficacy.
This study has some limitations. First, the sample size was relatively small, precluding meaningful subgroup analysis by lesion characteristics and lipiodol morphology. Second, the impact of portal vein tumor thrombus on the hemodynamics has not been systematically evaluated, which may alter tissue contrast media content and potentially affect tissue CT attenuation values on VNC images. Third, the differences in CT attenuation values among RVT, ANHP, and LRA on VNC images were not supplemented by subjective visual assessments.

5
Conclusion
VNC images can serve as a supplementary tool to conventional contrast-enhanced CT for efficacy evaluation in HCC patients after TACE by effectively reducing lipiodol and its associated artifacts, identifying RVT independent of enhancement characteristics, and directly visualizing tumor necrosis at lipiodol deposition sites.

Funding Statement
This work was supported by a grant from the 10.13039/501100001809National Natural Science Foundation of China (grant number: 82072003) for the collection and analysis of data.

[Funding and Acknowledgments]
This work was supported by a grant from the National Natural Science Foundation of China (grant number: 82072003) for the collection and analysis of data. The authors are grateful for the support from Dr. Jianying Li and for Dr. Wei Ren for their English expression advice during the drafting of this manuscript.

Ethical Statement
All procedures were performed in compliance with relevant laws and institutional guidelines and have been approved by the appropriate institutional committees. The ethical approval was obtained in July 2021, May 2024 and July 2024 (approval number: KY 2021–052–03, KY-2024–5–89–1, KYLL-2024B-039). Written informed consent was not required because written informed consent was obtained for undergoing the non-contrast and enhanced CT examination, but the informed consent for the post-analysis was waived in this study due to its retrospective nature.

CRediT authorship contribution statement
Mingxin Zhang: Writing – review & editing, Investigation. Kai Zhang: Writing – review & editing, Investigation. Liqin Zhao: Writing – review & editing, Supervision, Resources, Project administration, Funding acquisition. Taoming Du: Writing – review & editing, Writing – original draft, Methodology, Formal analysis, Conceptualization. Changchun Liu: Writing – review & editing, Investigation. Cheng Yan: Writing – review & editing, Formal analysis. Jingwen Zhang: Writing – review & editing, Formal analysis. Can Su: Writing – review & editing, Visualization, Validation. Jing Han: Writing – review & editing, Investigation. Yingxuan Wang: Writing – review & editing, Visualization, Validation. Jinghui Dong: Writing – review & editing, Supervision, Resources, Project administration. Mingzi Gao: Writing – review & editing, Writing – original draft, Methodology, Formal analysis, Conceptualization. Yujie Chen: Writing – review & editing, Visualization, Validation.

Declaration of Competing Interest
The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article.