CT is a commonly performed investigation for pediatric liver lesions. It is equivalent to MRI in staging the disease, including vascular evaluation [10]. Furthermore, CT chest for screening lung metastasis can also be performed in the same setting [10]. Although there is a potential risk related to radiation exposure on CT, it may get counterbalanced by the need for sedation or anesthesia for MRI [11].

The performance of CT/MRI in lesion resectability in pediatric liver tumors has been extensively studied [10, 12,13,14,15,16]. Though CT is commonly performed, studies assessing the accuracy of CT in diagnosing pediatric liver neoplasms are limited. A retrospective study in 2018 on 41 pediatric liver lesions (< 18-year-old patients) assessed the accuracy of CT and MRI in differentiating benign from malignant lesions based on Liver Reporting and Data System (LI-RADS) criteria. They showed a 100% (CI, 74–100%) sensitivity and 100% (CI, 81–100%) negative predictive value for diagnosing malignant liver lesions [17]. This study however did not diagnose the individual lesion pathology beyond labeling them benign or malignant, was not focused on the < 2 years age group, and encompassed seven hepatoblastomas, three hepatocellular carcinoma, and other predominant benign etiologies. No AFP correlation was performed [17].

Few other studies have evaluated the accuracy of imaging in specific diagnoses. A study evaluating focal liver lesions in children under 14 years in 45 patients found CT/MRI to be more sensitive and specific (95.2% and 90.2% respectively) in diagnosing hemangioma compared to ultrasound, when combined with clinical features like age, size of the lesion, and AFP values, with an AUC value of 0.93 (0.81–0.99) [6]. This is similar to our findings, with the classical enhancement pattern being pathognomonic for the diagnosis. Another 2018 study evaluating the accuracy of CT in diagnosing hepatoblastoma in 43 children (under 18 years age) showed an accuracy of 78.5% for tumors more than 4.5 cm and 71.4% for tumors < 4.5 cm on CT [18]. We observed a higher accuracy of 89.8% for our hepatoblastoma subset of patients on imaging without AFP correlation, which increased to 96.6% with AFP correlation.

There was a significant association between the presence of calcifications and hepatoblastomas, with 42% of hepatoblastomas in our study demonstrating calcifications. More than 50% of hepatoblastomas present with calcifications as per literature, and the commonly described patterns are coarse, chunky, and speckled types [19, 20]. The morphological patterns of calcifications we observed include focal and punctate, coarse/chunky, and curvilinear/rim calcifications. A few rare cases of predominantly calcified hepatoblastomas have also been reported, although we did not observe one in our cohort [21].

A hypo-attenuating central scar has been well established in FNH and fibrolamellar HCC but has also been described previously in hepatoblastoma [22]. Fibrolamellar HCC is not a differential diagnosis in this age group. FNH demonstrates intense homogeneous arterial enhancement and becomes isodense on the portal venous and delayed images, with the scar demonstrating delayed enhancement [23]. This was different from the hypo-enhancing scar-like appearance we observed in 25% of our cases, which significantly correlated with the diagnosis of hepatoblastoma. The pathological correlation for this morphology is not clear and needs further exploration. It could possibly represent central stroma between the tumor lobules. Larger studies are needed to validate this observation.

The diagnosis of rhabdoid on imaging alone was challenging. Aggressive features like metastasis, hemorrhage, necrosis, capsule rupture, or ascites without the typical morphology of hepatoblastoma were labeled indeterminate by the radiologists. The diagnosis was revised to hepatoblastoma or rhabdoid or retained as indeterminate after AFP correlation, given that rhabdoid tumors will not have raised AFP levels. Three cases with predominant cystic/necrotic components and metastases were accurately diagnosed as rhabdoid tumors after AFP levels, while the remaining three remained indeterminate.

PHITT guidelines state that if there are no differential diagnoses other than hepatoblastoma, a biopsy can be avoided after a multidisciplinary team discussion when the imaging and clinical findings are concordant [8]. In cases where the diagnosis is not certain or when a malignant lesion is suspected, the pediatric LI-RADS working group recommends a biopsy [24]. As per the German Society of Pediatric Hematology and Oncology (GPOH) in children below 3 years, a suspected case of hepatoblastoma with highly elevated AFP can be treated as hepatoblastoma without a biopsy [25]. This has been followed in many institutes since children less than 2 years are at higher risk for procedure or sedation/anesthesia related complications. Hemangiomas are also traditionally diagnosed on imaging alone without a biopsy in both children and adults.

With other differential diagnoses of a mesenchymal hamartoma and a rhabdoid tumor also present in the under 2-year population, we evaluated whether the lesions that are confidently diagnosed on imaging with AFP correlation could avoid a biopsy. If a policy of biopsying only indeterminate lesions based on imaging and AFP correlation was followed, a biopsy could be avoided in 93% lesions while retaining a high overall accuracy. This can be followed in lesions where the diagnosis is certain or when an upfront resection is possible. Overall, AFP correlation improved the accuracy of CT and decreased the number of indeterminate lesions, but this was not statistically significant. A larger multicentric study would be helpful to clarify these findings.

The strength of our study is that we analyzed data from a single institute with a good sample size of treatment naïve cases for assessing the accuracy. We specifically focused on the under 2-year population, which is seldom researched in isolation. The limitations of our study include a referral bias of being a tertiary oncology center, the retrospective nature of the study, and disproportionately higher cases of hepatoblastomas. We performed a consensus review with two radiologists rather than an independent analysis with interobserver agreement calculation, but we believe that this mirrors the practical scenario more accurately. We also did not evaluate the impact of multidisciplinary team discussions in the cases due to the retrospective nature of the study. Larger prospective studies would be required to evaluate this aspect.