LI-RADS Treatment Response Algorithm: Performance and Diagnostic Accuracy

Abstract

Background

In 2017, the Liver Imaging Reporting and Data System (LI-RADS) included an algorithm for the assessment of hepatocellular carcinoma (HCC) treated with local-regional therapy. The aim of the algorithm was to enable standardized evaluation of treatment response to guide subsequent therapy. However, the performance of the algorithm has not yet been validated in the literature.

Purpose

To evaluate the performance of the LI-RADS 2017 Treatment Response algorithm for assessing the histopathologic viability of HCC treated with bland arterial embolization.

Materials and Methods

This retrospective study included patients who underwent bland arterial embolization for HCC between 2006 and 2016 and subsequent liver transplantation. Three radiologists independently assessed all treated lesions by using the CT/MRI LI-RADS 2017 Treatment Response algorithm. Radiology and posttransplant histopathology reports were then compared. Lesions were categorized on the basis of explant pathologic findings as either completely (100%) or incompletely (<100%) necrotic, and performance characteristics and predictive values for the LI-RADS Treatment Response (LR-TR) Viable and Nonviable categories were calculated for each reader. Interreader association was calculated by using the Fleiss κ.

Results

A total of 45 adults (mean age, 57.1 years ± 8.2; 13 women) with 63 total lesions were included. For predicting incomplete histopathologic tumor necrosis, the accuracy of the LR-TR Viable category for the three readers was 60%–65%, and the positive predictive value was 86%–96%. For predicting complete histopathologic tumor necrosis, the accuracy of the LR-TR Nonviable category was 67%–71%, and the negative predictive value was 81%–87%. By consensus, 17 (27%) of 63 lesions were categorized as LR-TR Equivocal, and 12 of these lesions were incompletely necrotic. Interreader association for the LR-TR category was moderate (κ = 0.55; 95% confidence interval: 0.47, 0.67).

Conclusion

The Liver Imaging Reporting and Data System 2017 Treatment Response algorithm had high predictive value and moderate interreader association for the histopathologic viability of hepatocellular carcinoma treated with bland arterial embolization when lesions were assessed as Viable or Nonviable.

© RSNA, 2019

Online supplemental material is available for this article.

See also the editorial by Gervais in this issue.

Summary

The Liver Imaging Reporting and Data System, or LI-RADS, 2017 Treatment Response algorithm has high predictive value for the histopathologic viability of hepatocellular carcinoma treated with bland arterial embolization when observations are assessed as showing Viable or Nonviable disease, with moderate interreader association.

Key Points

• ■ Interreader association for final Liver Imaging Reporting and Data System (LI-RADS) Treatment Response (LR-TR) algorithm assessment category was moderate (κ = 0.55) and was similar to that for pre-embolization LI-RADS category association (κ = 0.57).

• ■ Absence of arterial phase hyperenhancement was the most important feature in the Treatment Response algorithm for predicting an LR-TR Nonviable lesion (ie, a lesion showing complete tumor necrosis after bland arterial embolization) (consensus area under the receiver operating characteristic curve, 0.69).

• ■ The majority of lesions assessed as LR-TR Equivocal were incompletely necrotic at histopathologic examination (12 [71%] of 17), suggesting that lesions in this category may warrant additional local-regional therapy.

Introduction

Hepatocellular carcinoma (HCC) is the most common primary liver malignancy and the second-most common cause of cancer-related death worldwide (1,2). For patients with stage T2 HCC who are not candidates for surgical resection, liver transplantation offers the best chance for disease-free survival (3). The limited availability of donor organs and increasing demand for these organs result in long waiting-list times, during which many patients become ineligible for transplantation because of tumor progression beyond stage T2 (4). This has prompted the widespread use of local-regional bridging therapies such as percutaneous ablation, transcatheter bland arterial embolization, chemoembolization, radioembolization, and radiation therapy to maintain a patient’s eligibility for transplantation (5,6).

The Liver Imaging Reporting and Data System (LI-RADS) is a system of standardized imaging criteria developed for the evaluation of patients at risk of developing HCC. The 2017 version of LI-RADS introduced a Treatment Response (LI-RADS Treatment Response [LR-TR]) algorithm for the assessment of lesions that have been previously treated with local-regional therapies (7,8). These lesions are categorized as LR-TR Nonviable, Equivocal, or Viable according to their imaging features, including focal arterial phase hyperenhancement (APHE), washout, and enhancement similar to pretreatment enhancement (9). While similar to the modified Response Evaluation Criteria in Solid Tumors (mRECIST) (10), the LR-TR algorithm clarifies tumor status at the level of individual lesions, rather than making an overall categorization of disease burden (11). However, there are currently no published data that evaluate the performance of the Treatment Response algorithm for predicting the degree of local-regional therapy–induced necrosis in individual lesions. Because the imaging assessment of treated HCC lesions is a key determinant of continued eligibility for liver transplantation and need for further therapy, understanding the diagnostic performance of the Treatment Response algorithm for assessing lesion viability is critical to making management decisions.

The purpose of this study was to evaluate the performance of the LI-RADS 2017 Treatment Response algorithm for assessing the viability of HCCs that have been treated with bland arterial embolization.

Materials and Methods

Our Institutional Review Board approved this retrospective study. The requirement for informed consent was waived. The study was compliant with the Health Insurance Portability and Accountability Act.

Study Population

A search of our institutional database was performed for consecutive patients with a history of HCC (according to imaging diagnosis), bland arterial embolization between 2006 and 2016, and liver transplantation (n = 93). Exclusion criteria were as follows: (a) local-regional therapy or other treatment prior to the first available multiphase CT or MRI examination (n = 21), (b) no multiphase CT or MRI examination prior to embolization (n = 1), and (c) no postembolization or pretransplant multiphase CT or MRI examination performed within 90 days before transplantation (n = 26). The 90-day interval was chosen on the basis of the maximum allowable interval between imaging and liver transplantation according to the Organ Procurement and Transplantation Network (OPTN) guideline (12). Imaging was required to be compliant with LI-RADS 2017 and OPTN technique recommendations, including the acquisition of adequately timed arterial phase images, for a patient’s inclusion in our study (11,13).

Radiology and histopathology reports were then reviewed and compared in consensus by a 3rd-year radiology resident (E.L.S.), an abdominal radiology faculty member with 9 years of post-fellowship experience (M.R.B.), and a liver pathology faculty member with expertise in liver transplantation and 9 years of post-fellowship experience (D.M.C.).

Clinical and laboratory data were collected from the electronic medical record, including age at embolization, sex, underlying etiology of chronic liver disease, number of days between embolization and transplant, Couinaud segment location of each embolized nodule, serum total bilirubin level, and Model for End-stage Liver Disease-Sodium, or MELD-Na, score (14).

Explant Histopathologic Examination

Histopathologic features for each treated lesion were collected from the electronic medical record, as assessed at the time of liver explantation by one of six liver pathologists with 5, 11, 12, 15, 20, and 30 years of experience. Histopathologic features collected included tumor location by Couinaud segment, tumor measurement in three dimensions, histopathologic diagnosis and tumor grade, and histopathologic tumor necrosis. Tumor necrosis was originally recorded as less than 50% necrosis, 50%–99% necrosis, or 100% necrosis and for the purpose of this work was consolidated to binary categories of less than 100% necrosis versus 100% necrosis. In cases where necessary histopathologic data in the reports were incomplete (for three of 63 lesions), a liver pathology team member (D.M.C.) performed a secondary review of the available slides to provide these data. This pathologist had access to the original histopathologic findings but was blinded to imaging findings and study reader assessments.

Diagnostic Imaging Technique

In patients with multiple CT or MRI examinations prior to embolization (39 of 45 patients), the LI-RADS 2017–compliant imaging study performed closest to the date of embolization was used. Variability in imaging modality was due to ordering provider preference. Technical details regarding the CT and MRI systems are described in Appendix E1 (online). Specific pulse sequence parameters for dynamic T1-weighted MRI are summarized in Table E1 (online).

Image Analysis

Two faculty abdominal radiologists with 11 and 17 years of abdominal MRI experience (B.C.A. [reader 1] and C.M.M. [reader 2], respectively) and one faculty interventional radiologist with 5 years of abdominal MRI experience (J.R. [reader 3]) independently evaluated each patient’s pre- and postembolization imaging studies and assessed all lesions by using the LI-RADS 2017 CT/MRI (pre-embolization) and Treatment Response (postembolization) algorithms. All readers were blinded to clinical information and radiologic reports. On pre-embolization images, readers assessed lesion size and the presence of LI-RADS major features, including APHE, washout, capsule, and threshold growth (when prior imaging data were available). Pre-embolization LI-RADS category was assigned based on major features only. On postembolization/pretransplant images, readers were asked to assess lesion visibility, Treatment Response category, presence or absence of individual Treatment Response features, and sizes of Viable or Equivocal disease.

Lesions classification was defined by LI-RADS 2017; schematic representations of these categories can be found in the CT/MRI LI-RADS 2017 document (7). In brief, LR-TR Nonviable lesions had expected posttreatment lesion hyperenhancement. Lesions that were no longer visible after treatment or had none of the LR-TR Viable features were also categorized LR-TR Nonviable lesions. LR-TR Viable lesions had nodular, mass-like, or irregular thick tissue in or along the treated lesion with APHE, washout, and/or enhancement similar to that before embolization. LR-TR Equivocal lesions had indeterminate enhancement and did not meet criteria for being probably or definitely viable.

In cases of reader uncertainty between two categories, a tiebreaking rule was applied, per Treatment Response algorithm guidelines, to choose the category reflecting lower certainty (ie, lesions with lower reader certainty of viability or nonviability were categorized as Equivocal) (7).

Statistical Analysis

Statistical analyses were performed by using open-source R statistical software (version 3.4.3 [2017]; the R Foundation for Statistical Computing, Vienna, Austria). The LR-TR Viable and Nonviable categories were reduced to binary indicators (compared against LR-TR Nonviable plus Equivocal and LR-TR Viable plus Equivocal categories, respectively) for the calculation of accuracy, positive predictive value, and negative predictive value of the LR-TR Viable assessment category to predict less than 100% histopathologic necrosis (incomplete necrosis) and the ability of the LR-TR Nonviable assessment category to predict 100% histopathologic necrosis (complete necrosis). Interreader association for discrete variables was assessed by using the Fleiss κ, with 95% confidence intervals (CIs) obtained by bootstrapping. For the κ value, association was interpreted as slight (0.01–0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), or almost perfect (0.81–1.00) (15).

The area under the receiver operating characteristic curve (AUC) was used to summarize the ability of independent variables, including smaller postembolization Viable or Equivocal size, absence of imaging features of residual tumor, and LR-TR assessment category to predict the outcome of complete necrosis (100% necrosis vs <100% necrosis).

Results

Our final study population comprised 45 adults (mean age at embolization, 57.1 years ± 8.2; range, 27–70 years) with 63 total lesions (Fig 1). This included 13 women (mean age, 55.6 years ± 9.8) and 32 men (mean age, 57.8 years ± 7.5). Study population characteristics prior to embolization are summarized in Table 1. The 45 pre-embolization examinations included 19 CT examinations, 19 MRI examinations with gadobenate dimeglumine, and seven MRI examinations with gadoxetate disodium. The 45 postembolization/pretransplantation examinations included 24 CT studies, 18 MRI studies with gadobenate dimeglumine, and three MRI studies with gadoxetate disodium. Of the 45 patients, 17 had both pre- and postembolization imaging performed with CT, 19 had both performed with MRI, and nine had imaging with a combination of the two modalities.

Figure 1:

Study flowchart shows a summary of the inclusion and exclusion criteria and final study population.

Table 1:

Study Population Characteristics Prior to Bland Arterial Embolization

Note.—APHE = arterial phase hyperenhancement, LI-RADS = Liver Imaging Reporting and Data System, NASH = nonalcoholic steatohepatitis.

* Data in parentheses are ranges.

† Data are numbers of patients or numbers of lesions, with percentages in parentheses.

A total of 63 lesions were assigned to the following pre-embolization LI-RADS categories based on the consensus of two of three readers: LI-RADS 4 (n = 16), LI-RADS 5 (n = 45), and LI-RADS TIV (n = 2). The 16 LI-RADS 4 lesions were either empirically embolized following characterization as hypervascular at arteriography or were embolized as part of a lobar embolization for coexisting LI-RADS 5 lesion(s). The majority of lesions had a single embolization treatment. Nine of 63 lesions had two separate embolizations. Time between embolization and transplant was not taken into account in the statistical model (as detailed in Appendix E1 [online]).

Representative examples of lesions rated LR-TR Viable, LR-TR Nonviable, and LR-TR Equivocal by consensus are presented in Figures 2–4.

Figure 2a:

Images in 65-year-old man with cirrhosis due to nonalcoholic steatohepatitis. (a, b) Images obtained at 3.0-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 4.2/1.2; flip angle, 15°; matrix, 320 × 248). Axial (a) arterial and (b) portal venous phase images obtained before embolization show a 42-mm hyperenhancing lesion (arrow) with washout that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (4.1/2.0; flip angle, 10°; matrix, 256 × 205). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 71 days after embolization, the lesion (arrow) had increased in size to 51 mm. It demonstrated mass-like irregular arterial phase hyperenhancement. This lesion was categorized as LR-TR Viable by all three readers. At histologic examination, this lesion was less than 50% necrotic.

Figure 4a:

Images in 60-year-old woman with cirrhosis due to chronic hepatitis C infection. (a, b) Images obtained at 1.5-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 7.5/2.4; flip angle, 10°; matrix, 256 × 170). Axial (b) arterial and (b) portal venous phase images obtained before embolization show a 21-mm hyperenhancing lesion (arrow) with washout that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (7.5/2.4; flip angle, 10°; matrix, 256 × 170). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 168 days after embolization, the lesion (arrow) had decreased in size to 7 mm. Two of three readers assessed indeterminate arterial phase hyperenhancement, while one reader assessed focal mass-like arterial hyperenhancement. This lesion was categorized as LR-TR Equivocal by two readers and as LR-TR Viable by one reader. At histologic examination, this lesion was less than 50% necrotic.

Figure 2b:

Images in 65-year-old man with cirrhosis due to nonalcoholic steatohepatitis. (a, b) Images obtained at 3.0-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 4.2/1.2; flip angle, 15°; matrix, 320 × 248). Axial (a) arterial and (b) portal venous phase images obtained before embolization show a 42-mm hyperenhancing lesion (arrow) with washout that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (4.1/2.0; flip angle, 10°; matrix, 256 × 205). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 71 days after embolization, the lesion (arrow) had increased in size to 51 mm. It demonstrated mass-like irregular arterial phase hyperenhancement. This lesion was categorized as LR-TR Viable by all three readers. At histologic examination, this lesion was less than 50% necrotic.

Figure 2c:

Images in 65-year-old man with cirrhosis due to nonalcoholic steatohepatitis. (a, b) Images obtained at 3.0-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 4.2/1.2; flip angle, 15°; matrix, 320 × 248). Axial (a) arterial and (b) portal venous phase images obtained before embolization show a 42-mm hyperenhancing lesion (arrow) with washout that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (4.1/2.0; flip angle, 10°; matrix, 256 × 205). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 71 days after embolization, the lesion (arrow) had increased in size to 51 mm. It demonstrated mass-like irregular arterial phase hyperenhancement. This lesion was categorized as LR-TR Viable by all three readers. At histologic examination, this lesion was less than 50% necrotic.

Figure 2d:

Images in 65-year-old man with cirrhosis due to nonalcoholic steatohepatitis. (a, b) Images obtained at 3.0-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 4.2/1.2; flip angle, 15°; matrix, 320 × 248). Axial (a) arterial and (b) portal venous phase images obtained before embolization show a 42-mm hyperenhancing lesion (arrow) with washout that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (4.1/2.0; flip angle, 10°; matrix, 256 × 205). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 71 days after embolization, the lesion (arrow) had increased in size to 51 mm. It demonstrated mass-like irregular arterial phase hyperenhancement. This lesion was categorized as LR-TR Viable by all three readers. At histologic examination, this lesion was less than 50% necrotic.

Figure 3a:

Images in 68-year-old man with cirrhosis secondary to chronic hepatitis C infection. (a, b) Images obtained at 3.0-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 4.2/1.3; flip angle, 15°; matrix, 320 × 256). Axial (a) arterial and (b) portal venous phase images obtained before embolization show a 13-mm hyperenhancing lesion (arrow) with washout and a capsule that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (4.4/2.1; flip angle, 12°; matrix, 256 × 192). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 227 days after embolization, there was no residual arterial phase hyperenhancement or washout associated with the lesion (arrow). This lesion was categorized as LR-TR Nonviable by all three readers. At histologic examination, this lesion was 100% necrotic.

Figure 3b:

Images in 68-year-old man with cirrhosis secondary to chronic hepatitis C infection. (a, b) Images obtained at 3.0-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 4.2/1.3; flip angle, 15°; matrix, 320 × 256). Axial (a) arterial and (b) portal venous phase images obtained before embolization show a 13-mm hyperenhancing lesion (arrow) with washout and a capsule that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (4.4/2.1; flip angle, 12°; matrix, 256 × 192). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 227 days after embolization, there was no residual arterial phase hyperenhancement or washout associated with the lesion (arrow). This lesion was categorized as LR-TR Nonviable by all three readers. At histologic examination, this lesion was 100% necrotic.

Figure 3c:

Images in 68-year-old man with cirrhosis secondary to chronic hepatitis C infection. (a, b) Images obtained at 3.0-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 4.2/1.3; flip angle, 15°; matrix, 320 × 256). Axial (a) arterial and (b) portal venous phase images obtained before embolization show a 13-mm hyperenhancing lesion (arrow) with washout and a capsule that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (4.4/2.1; flip angle, 12°; matrix, 256 × 192). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 227 days after embolization, there was no residual arterial phase hyperenhancement or washout associated with the lesion (arrow). This lesion was categorized as LR-TR Nonviable by all three readers. At histologic examination, this lesion was 100% necrotic.

Figure 3d:

Images in 68-year-old man with cirrhosis secondary to chronic hepatitis C infection. (a, b) Images obtained at 3.0-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 4.2/1.3; flip angle, 15°; matrix, 320 × 256). Axial (a) arterial and (b) portal venous phase images obtained before embolization show a 13-mm hyperenhancing lesion (arrow) with washout and a capsule that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (4.4/2.1; flip angle, 12°; matrix, 256 × 192). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 227 days after embolization, there was no residual arterial phase hyperenhancement or washout associated with the lesion (arrow). This lesion was categorized as LR-TR Nonviable by all three readers. At histologic examination, this lesion was 100% necrotic.

Figure 4b:

Images in 60-year-old woman with cirrhosis due to chronic hepatitis C infection. (a, b) Images obtained at 1.5-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 7.5/2.4; flip angle, 10°; matrix, 256 × 170). Axial (b) arterial and (b) portal venous phase images obtained before embolization show a 21-mm hyperenhancing lesion (arrow) with washout that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (7.5/2.4; flip angle, 10°; matrix, 256 × 170). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 168 days after embolization, the lesion (arrow) had decreased in size to 7 mm. Two of three readers assessed indeterminate arterial phase hyperenhancement, while one reader assessed focal mass-like arterial hyperenhancement. This lesion was categorized as LR-TR Equivocal by two readers and as LR-TR Viable by one reader. At histologic examination, this lesion was less than 50% necrotic.

Figure 4c:

Images in 60-year-old woman with cirrhosis due to chronic hepatitis C infection. (a, b) Images obtained at 1.5-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 7.5/2.4; flip angle, 10°; matrix, 256 × 170). Axial (b) arterial and (b) portal venous phase images obtained before embolization show a 21-mm hyperenhancing lesion (arrow) with washout that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (7.5/2.4; flip angle, 10°; matrix, 256 × 170). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 168 days after embolization, the lesion (arrow) had decreased in size to 7 mm. Two of three readers assessed indeterminate arterial phase hyperenhancement, while one reader assessed focal mass-like arterial hyperenhancement. This lesion was categorized as LR-TR Equivocal by two readers and as LR-TR Viable by one reader. At histologic examination, this lesion was less than 50% necrotic.

Figure 4d:

Images in 60-year-old woman with cirrhosis due to chronic hepatitis C infection. (a, b) Images obtained at 1.5-T MRI performed by using a three-dimensional (3D) T1-weighted gradient-recalled echo sequence (repetition time msec/echo time msec, 7.5/2.4; flip angle, 10°; matrix, 256 × 170). Axial (b) arterial and (b) portal venous phase images obtained before embolization show a 21-mm hyperenhancing lesion (arrow) with washout that was deemed a Liver Imaging Reporting and Data System, or LI-RADS, category 5 lesion. (c, d) Images obtained at 1.5-T MRI performed by using a 3D T1-weighted gradient-recalled echo sequence (7.5/2.4; flip angle, 10°; matrix, 256 × 170). On pretransplant axial (c) arterial and (d) portal venous phase images obtained 168 days after embolization, the lesion (arrow) had decreased in size to 7 mm. Two of three readers assessed indeterminate arterial phase hyperenhancement, while one reader assessed focal mass-like arterial hyperenhancement. This lesion was categorized as LR-TR Equivocal by two readers and as LR-TR Viable by one reader. At histologic examination, this lesion was less than 50% necrotic.

Diagnostic Performance for Predicting Histologic Tumor Necrosis

Reader performance for predicting incomplete necrosis and complete necrosis is summarized in Table 2. Reader accuracy for predicting incomplete necrosis based on a determination of LR-TR Viable disease ranged from 60% to 65% between readers. Accuracy for predicting complete necrosis based on a determination of LR-TR Nonviable disease ranged from 67% to 71%. The positive predictive value for predicting incomplete necrosis by using the LR-TR Viable category ranged from 86% to 96%. The negative predictive value for predicting complete necrosis by using the LR-TR Nonviable category ranged from 81% to 87%. By consensus, 17 (27%) of the 63 lesions were categorized as LR-TR Equivocal, and 71% of these (12 of 17) were incompletely necrotic at histologic examination. Postembolization LR-TR Equivocal lesion characteristics, including percentage necrosis, major imaging features, and mean whole lesion and Viable disease sizes are summarized in Table 3. Analysis of reader performance for predicting histologic tumor necrosis in the subpopulation of LR-5 and LR-TIV nodules (n = 47) was also performed, demonstrating results similar to those for the total lesion population (Table 2). By consensus, 28% (13 of 47) of LR-5 and LR-TIV lesions were categorized as LR-TR Equivocal, and 69% of these lesions (nine of 13) were incompletely necrotic at histologic examination.

Table 2:

Reader Performance for Predicting Incomplete Necrosis (Reader Classification as LR-TR Viable) and Complete Necrosis (Reader Classification as LR-TR Nonviable)

Note.—Data are numbers of lesions, with diagnostic values as percentages in parentheses and 95% confidence intervals in brackets. The majority of patients had a single lesion; however, a small number of patients (12 of 45) had more than one lesion. The total lesion population (n = 63) includes those patients with multiple lesions. LI-RADS = Liver Imaging Reporting and Data System, NPV = negative predictive value, PPV = positive predictive value.

Table 3:

Postembolization Lesion Characteristics by Consensus LR-TR Algorithm Category

Note.—Unless otherwise specified, data are numbers of lesions. APHE = arterial phase hyperenhancement; LI-RADS = Liver Imaging Reporting and Data System; LR-TR = LI-RADS Treatment Response; NMLIT = nodular, mass-like, or irregular thick tissue in or along the treated lesion.

* Data are means ± standard deviations.

A small number of patients (12 of 45) had more than one lesion. Analysis restricting the data set to a single lesion per patient was performed to assess within-patient bias and demonstrated no substantial difference in performance (Table E2 [online]).

Interreader Association

Interreader association for pre- and postembolization imaging features is summarized in Table E3 (online). Interreader association for pre-embolization LI-RADS category was moderate (κ = 0.57; 95% CI: 0.44, 0.71). For postembolization features, association was substantial for lesion visibility (κ = 0.75; 95% CI: 0.52, 1.0) and for the presence of nodular, mass-like, or irregular thick tissue in or along the treated lesion with APHE (κ = 0.66; 95% CI: 0.53, 0.78). Association was variable for other features and was fair for indeterminate enhancement (κ = 0.25; 95% CI: 0.11, 0.40). Interreader association for final LR-TR assessment category was moderate (κ = 0.55; 95% CI: 0.47, 0.67) and similar to pre-embolization LI-RADS categorization association.

Association between Imaging Features and Histopathologic Tumor Necrosis

Final LR-TR assessment category was associated with AUCs of 0.66–0.76 for predicting complete necrosis (Table 4). For predicting complete necrosis, smaller postembolization Viable or Equivocal size and absence of APHE were associated with the highest AUCs, which ranged from 0.66 to 0.83. The presence of expected posttreatment enhancement and absence of washout were associated with AUCs of 0.52–0.69. Absence of enhancement similar to pretreatment appearance and absence of indeterminate enhancement were associated with the lowest AUCs, which ranged from 0.42 to 0.62.

Table 4:

Results of AUC Analysis of Size, Imaging Feature, and LR-TR Assessment Category Performance for Predicting Histopathologic Necrosis in 63 Lesions

Note.—Data in parentheses are 95% confidence intervals. APHE = arterial phase hyperenhancement; AUC = area under the receiver operating characteristic curve; LI-RADS = Liver Imaging Reporting and Data System; NMLIT = nodular, mass-like, or irregular thick tissue in or along the treated lesion.

Discussion

Our results show that for lesions treated with bland embolization that meet criteria for either the LR-TR Nonviable or Viable categories, the 2017 version of the Liver Imaging Reporting and Data System (LI-RADS) Treatment Response (LR-TR) algorithm performs well for predicting complete necrosis (LR-TR Nonviable category: negative predictive value = 81%–87%; accuracy = 67%–71%) and incomplete necrosis (LR-TR Viable category: positive predictive value = 86%–96%; accuracy = 60%–65%). However, a substantial proportion (17 [27%] of 63) of embolized lesions could not be definitively characterized and fell into the LR-TR Equivocal category. Of those lesions rated LR-TR Equivocal, most (12 [71%] of 17) were incompletely necrotic at histopathologic examination. This suggests that in patients who might benefit from additional local-regional therapies, LR-TR Equivocal lesions may warrant close imaging follow-up or further treatment. Importantly, the performance of the Treatment Response algorithm was similar when we considered a mixed population of treated LR-4, LR-5, and LR-TIV lesions versus definite HCC lesions (only LR-5 and LR-TIV lesions).

One of the primary goals of standardized reporting systems like LI-RADS is to improve interreader association in reporting, an often challenging task when lesion categorization can be subjective. In our study, a small number of LR-TR Equivocal lesions were assessed by consensus as having features of the LR-TR Viable category (focal APHE and/or washout) or the LR-TR Nonviable category (expected posttreatment lesional hyperenhancement). We believe this reflects reader uncertainty with the appropriate use of the criteria for LR-TR Viable and Nonviable response categories and application of the Treatment Response algorithm tiebreaking rule guidelines to choose the category reflecting lower certainty. Because the majority of LR-TR Equivocal lesions were found to be incompletely necrotic at histopathologic examination, strict interpretation of the Treatment Response algorithm may have produced an appropriate reassignment of some of the incompletely necrotic lesions to the LR-TR Viable category. While continued improvement is needed, we observed comparable moderate interreader association between the final categorization of treated lesions (κ = 0.55) and untreated lesions (κ = 0.57). Importantly, interreader association was particularly low for the “indeterminate hyperenhancement” component of the Treatment Response algorithm, and absence of both “indeterminate hyperenhancement” and “enhancement similar to pretreatment” demonstrated low predictive performance (consensus AUC = 0.53 for each for predicting complete necrosis). Interreader association for “enhancement similar to pretreatment” was not calculated because of a small sample size. The dominant posttreatment imaging feature of the LR-TR Equivocal lesions was “indeterminate hyperenhancement,” and, while a low AUC for indeterminate nodules may be expected, consideration should be given to simplifying or clarifying this part of the definition to improve interreader association.

A number of systems have been described for standardized measurement and reporting of treatment response in solid tumors (eg, the Response Evaluation Criteria in Solid Tumors [RECIST], the mRECIST, the Choi criteria, and the modified Choi criteria) that are intended for evaluation of treatment response at the patient level (9,16). The LR-TR algorithm is distinct in that it focuses on lesion-level assessments rather than on patient-level assessments. This is particularly important in the assessment of HCC, because patients may develop multiple lesions over time, necessitating different local-regional therapies for maintenance of transplant eligibility or disease control. Lesion response within a single organ can be heterogeneous, requiring evaluation at the lesion-specific level.

LI-RADS 2017 defines posttreatment viability as the presence of live tumor cells within or along the margin of a treated lesion (7), recognizing that imaging is not sensitive for the detection of microscopic or small foci of residual tumor. The specific imaging features associated with tumor viability are derived from the major imaging features required for classification of an untreated lesion as LR-5, which have previously been shown to demonstrate moderate concordance between readers (17,18). The APHE feature is emphasized in other systems such as mRECIST, and our study showed that this feature was both the feature most strongly associated with viability and the postembolization feature with the greatest interreader association. One key difference between the LR-TR algorithm and other systems such as mRECIST is the inclusion of the washout feature. We noted that postembolization washout was never observed independently of APHE. Also, the incorporation of multiple features into the final LR-TR category only marginally improved AUCs compared with APHE alone.

Our study had several limitations, including its single-center retrospective design. All patients were evaluated and treated at a tertiary hospital and transplant center, resulting in biased patient selection that limits the generalizability of our results to patients treated at nontransplant centers. The use of histopathologic necrosis as the reference standard for Treatment Response algorithm category accuracy may have biased our performance results because of the inherent inability of imaging to depict microscopic or small foci of residual tumor. We also acknowledge that ultimately, patient outcomes would be a better metric for assessing Treatment Response algorithm accuracy and utility. The inclusion of different diagnostic imaging modalities introduced heterogeneity but could be viewed as a strength given that this heterogeneity reflects clinical practice. We were unable to explore differences in performance of the Treatment Response algorithm based on diagnostic modality because of the relatively small numbers of examinations performed with each modality individually. The time span of the study (2006–2016) is also a limitation; however, our institutional imaging protocol for HCC, including MRI protocols and technical capabilities, did not substantially change during that time. Also, our study population included only patients with a history of bland arterial embolization. A variety of local-regional therapies are currently available for treatment of HCC, and each result in different characteristic posttreatment appearances (9). Further investigations are needed, to determine both the generalizability of our results regarding bland arterial embolization and the performance of the Treatment Response algorithm after other types of local-regional therapy.

In conclusion, our study shows that the Liver Imaging Reporting and Data System (LI-RADS) 2017 Treatment Response algorithm performs well for hepatocellular carcinoma treated with bland arterial embolization when lesions are assessed as Viable or Nonviable disease. Arterial phase hyperenhancement was the most important feature of viability, suggesting that the Treatment Response algorithm may not outperform mRECIST on a lesional basis given their overlap, despite the greater complexity of the Treatment Response algorithm. Further research including direct comparison between the two algorithms is needed to determine whether the added complexity of the Treatment Response algorithm provides any significant benefit.