Liver Imaging Reporting and Data System in Patients at High Risk for Hepatocellular Carcinoma in the Memphis Veterans Affairs Population

Radiologist 1 identified 11 patients with LR4 and 21 patients with LR5. His scores showed a sensitivity of 64.3% and specificity of 82.1% with accuracy of 71.4% for LI-RADS in identifying eventual HCC. The false-negative rate of the LI-RADS diagnostic algorithm for radiologist 1 was 35.7% and the false-positive rate was 17.9%. Radiologist 2 identified 17 patients LR4 and 19 patients with LR5. Radiologist 2’s scores showed a sensitivity of 69.0% and specificity of 75.0% with accuracy of 71.4% for LI-RADS in identifying eventual HCC.The false-negative rate of the LI-RADS diagnostic algorithm for radiologist 2 was 31.0% and false-positive rate of 25.0%. The κ statistic was calculated to determine the interrater agreement. The radiologists agreed on 58 of 70 samples; 15 without HCC and 43 with HCC. The κ statistic was 0.592, which indicates moderate agreement (Table 2).

Discussion

If HCC is diagnosed late in the disease process based on symptomatology and not on surveillance imaging, the likelihood of receiving early and potential curative therapy greatly declines as was shown in a systemic literature review.⁹ Surveillance imaging and lesion interpretation by various radiologists has been difficult to standardize as new technologic advances continue to occur in the imaging of HCC.¹⁴ LI-RADS was initiated to help standardize CT and MRI interpretation and reporting of hepatic nodules. As a dynamic algorithm, it continues to adjust with new advances in imaging techniques with the most recent updates being made to the algorithm in 2014.^14,19 LI-RADS applies to patients at high risk for HCC most often who are already enrolled in a surveillance program.¹⁹ The MVAMC has a high incidence of patients with cirrhosis who are at risk for HCC, which is why we chose it as our study population.

LI-RADS can be applied to both MRI and CT imaging. Much of the recent literature have looked at LI-RADS in terms of MRI. A group in China looked at 100 pathologically confirmed patients and assigned a LI-RADS score to the MRI at the time of diagnosis and showed that MRI LI-RADS scoring was highly sensitive and specific in the diagnosis of HCC.²⁰ This study did note a numeric difference in the specificity of LI-RADS algorithm depending on how LR3 scores were viewed. If a LR3 score was considered negative rather than positive for HCC, then the specificity increased by almost 20%.²⁰

Another study looked at patients with liver nodules ≤ 20 mm found on ultrasound and obtained MRIs and biopsies on these patients, assigning the MRI a LI-RADs score.¹⁷ Darnell and colleagues found that MRI LR4 and LR5 have a high specificity for HCC. However, 29 of the 42 LR3 lesions examined were found to be HCC.¹⁷ Furthermore, Choi and colleagues retrospectively looked at patients in a HCC surveillance program who had undergone MRI as part of the program and assigned LI-RADS scores to these MRIs.²¹ Their study showed that LR5 criteria on gadoxetate disodium-enhanced MRI has excellent positive predictive value (PPV) for diagnosing HCC, and LR4 showed good PPV.²¹

In our study, we chose to look at LI-RADS in terms of surveillance CT scans 6 to 13 months prior to the diagnosis of HCC to see whether this method would allow us to intervene earlier with more aggressive diagnostics or therapy in those suspected of having HCC. Although Choi and colleagues looked retrospectively at MRI surveillance imaging, most of the prior studies have looked at LI-RADS scoring in imaging at the time of diagnosis.^17,20,21 By looking at surveillance CT scans, we sought to determine LI-RADS sensitivity, specificity, and diagnostic accuracy as a screening tool compared with CT evaluations without LI-RADS scoring.

We also chose to look at CT scans since most of the prior studies have looked at the more detailed and often more expensive MRIs. For both radiologists 1 and 2, the sensitivity was > 60% and specificity was > 70% with a diagnostic accuracy of 71.4% in predicting a diagnosis of HCC in future scans. Although there was high false negative of > 30% for both radiologists, we did consider LR3 as negative for HCC. As Darnell and colleagues’ study of MRI LI-RADS shows, LR3 may need to be revised in the future as its ambiguity can lead to false-negatives.¹⁷ Our results also showed moderate interreader agreement, which has been seen in previous studies with LI-RADS.¹⁸

Some studies have compared MRI with CT imaging in terms of LI-RADs classification of hepatic nodules to find out whether concordance was seen.^22,23 Both studies found that there was substantial discordance between MRI and CT with CT often underscoring hepatic nodules.^22,23 In Zhang and colleagues, interclass agreement between CT and MRI varied the most in terms of arterial enhancement with CT producing false-negative findings.²² CT also underestimated LI-RADS score by 16.9% for LR3, 37.3% for LR4, and 8.5% for LR5 in this study.²² Furthermore, Corwin and colleagues found a significant upgrade in terms of LI-RADS categorization with MRI for 42.5% of observations.²³ In this study, upgraded LI-RADS scores on MRI included 2 upgraded to LR5V (Figure), 15 upgraded to LR5, and 12 upgraded to LR4.²³

Our study shows that the LI-RADS algorithm has a good sensitivity, specificity, and diagnostic accuracy as a screening tool, predicting HCC in scans earlier than standard CT evaluation. In our study, the patients with HCC were shown to have higher LI-RADS scores on prediagnostic imaging, while the benign liver nodule patients were shown to have lower LI-RADS scores. This data would suggest that a LI-RADS score given to surveillance CT of LR4 or higher should recommend either a biopsy or follow-up imaging after a short interval. If LI-RADS is applied to surveillance CTs in patients at risk for HCC, a diagnosis of HCC may be arrived at earlier as compared with not using the LI-RADS algorithm. Earlier detection may lead to earlier intervention and improved treatment outcomes.