That system is LI-RADS, the Liver Imaging Reporting and Data System, introduced by the American College of Radiology in 2011 as a standardized lexicon and diagnostic algorithm for computed tomography and magnetic resonance imaging of the liver. Its great achievement was consistency. By defining a shared vocabulary of imaging features and a deterministic category assignment ranging from LR-1, meaning definitely benign, through LR-5, meaning definitely hepatocellular carcinoma, LI-RADS allowed radiologists worldwide to communicate findings in a common language. The LR-5 category was so well validated that major clinical guidelines accepted it as grounds for definitive noninvasive diagnosis, sparing many cirrhotic patients a biopsy altogether. But as Lee and Wei lay out in their commentary, a paradigm shift is underway. Mounting evidence from multicenter studies shows that LI-RADS categories and their constituent imaging features are powerful independent predictors of survival, recurrence, and treatment response, effectively transforming a static diagnostic schema into a dynamic prognostic biomarker that captures disease aggressiveness without a single needle entering the patient.

The biological logic behind this prognostic power is what makes the story compelling to oncologists and radiologists alike. Imaging features are not arbitrary visual patterns; they are the visible fingerprints of pathomolecular characteristics. The LR-M category, assigned to lesions that are probably or definitely malignant but not imaging-specific for hepatocellular carcinoma, consistently associates with poorer overall survival and recurrence-free survival compared with LR-4 or LR-5 lesions, even when histopathology confirms the tumor is indeed hepatocellular carcinoma. Pathomolecular analyses reveal why: LR-M tumors more frequently harbor microvascular invasion, poor differentiation, and progenitor cell markers, all hallmarks of aggressive biology. The LR-TIV category, which provides an unambiguous imaging diagnosis of macroscopic vascular invasion, is universally accepted as a marker of advanced disease and a strong negative prognostic factor. In other words, the radiologist’s report already contains a molecular biopsy, if only clinicians learn to read it that way.

Specific features map onto specific phenotypes with striking precision. Rim arterial phase hyperenhancement, a hallmark LR-M feature, is associated with the proliferative molecular class of hepatocellular carcinoma and with the macrotrabecular-massive histologic subtype, both linked to aggressive behavior. Non-smooth or infiltrative margins serve as imaging surrogates for microvascular invasion and aggressive histologic subtypes, consistently predicting increased risk of recurrence after surgery. Peritumoral enhancement in the arterial or early portal venous phase, known as corona enhancement, and peritumoral hypointensity on hepatobiliary phase images reflect disturbances in peritumoral hemodynamics and associate with microvascular invasion and dense CD8-positive T-cell infiltration, predicting higher early recurrence rates and shorter time to recurrence after curative-intent therapy. On the favorable side of the ledger, intratumoral fat and hepatobiliary phase hyperintensity correlate with well-differentiated tumors and non-proliferative molecular classes, suggesting a more indolent biology.

Yet the commentary emphasizes a crucial nuance that prevents the story from becoming a simple good-versus-bad narrative. Hepatobiliary phase hyperintense tumors, despite being typically well differentiated, frequently exhibit an immune-excluded tumor microenvironment, often driven by activation of the Wnt/beta-catenin pathway, and have been linked to poor response to immune checkpoint inhibitors. A tumor that looks biologically favorable in one therapeutic context may be resistant to the very immunotherapy a patient is about to receive. Conversely, intratumoral steatosis has been associated with immune exhaustion and, intriguingly, with favorable response to the atezolizumab plus bevacizumab combination in nonviral hepatocellular carcinoma. The lesson, the authors stress, is that imaging biomarkers do not indicate good or bad prognosis in absolute terms; they reflect specific underlying biology that may confer resistance or sensitivity to particular treatments, and LI-RADS features must therefore be interpreted within the specific treatment context to guide personalized decision-making.

This treatment-context principle has concrete consequences across every stage of disease. For patients with early-stage tumors eligible for curative-intent therapy, imaging-based risk stratification directly informs surgical planning. An LR-M classification or features predicting microvascular invasion, such as a nonsmooth margin or peritumoral enhancement, favor surgical resection with wider margins over ablation, and suggest the need for anatomic resection rather than a limited wedge excision. In the transplantation setting, the LR-M category and peritumoral hepatobiliary phase hypointensity on preoperative MRI independently predict recurrence even among patients who fall within the Milan criteria, raising the possibility of refining donor allocation with imaging-derived risk signals that currently go unreported.

For intermediate and advanced disease, the stakes are equally high. Transarterial chemoembolization is the standard recommendation for intermediate-stage tumors and systemic therapy for advanced disease, but considerable variability exists within these broad categories. Very large, multifocal, or bilobar tumors often respond poorly to chemoembolization, and several prognostic models already incorporate imaging-derived tumor burden scores to predict survival. Beyond size and number, tumor phenotype matters: hepatocellular carcinomas with imaging features of the proliferative subtype, such as infiltrative enhancement and necrosis, tend to respond poorly to chemoembolization. Identifying such cases on pretreatment imaging allows clinicians to redirect patients sooner toward systemic therapy rather than subjecting them to multiple rounds of an ineffective locoregional procedure. On the systemic side, hepatobiliary phase hyperintensity has been associated with worse progression-free survival following immune checkpoint inhibitor-based therapies, likely reflecting Wnt/beta-catenin-driven immune exclusion, while quantitative radiomic models have shown promise in predicting response to tyrosine kinase inhibitors, including a recent multicenter study that used pretreatment CT to predict which patients would respond to lenvatinib, allowing non-responders to be triaged to other options.

A notable recent development underscoring this evolution is the LI-RADS v2024 Treatment Response Assessment update, which introduces a separate assessment pathway for radiation-based therapies such as transarterial radioembolization. Post-treatment inflammation after radioembolization can mimic viable tumor on conventional scans, and the update addresses this by replacing the ambiguous LR-TR Equivocal category with LR-TR Nonprogressing and refining the criteria for LR-TR Viable, improving correlation with overall survival. The authors argue that the logical next step is a dedicated treatment response module for systemic therapies, one capable of handling the unique patterns of response and failure seen with targeted agents and immunotherapies, where apparent tumor enlargement can reflect pseudoprogression rather than true growth. The prognostic utility of the system also extends beyond CT and MRI to contrast-enhanced ultrasound, where early and marked washout within sixty seconds or rim arterial phase hyperenhancement, hallmark LR-M features, correlate with microvascular invasion, poor differentiation, and the proliferative molecular subclass, giving even resource-limited settings a real-time window into tumor biology.

The path forward, according to the commentary, runs through artificial intelligence and through a change in reporting culture. Interobserver variability in assessing features such as nonsmooth margins or rim enhancement remains a real limitation, and AI-driven radiomics offers objective, quantitative feature extraction that could standardize risk stratification. Deep learning models already achieve sensitivity and specificity above ninety percent for LR-5 detection, and radiomic signatures predict microvascular invasion, histologic grade, and treatment response more accurately than visual assessment alone, while integrated models combining radiomic features, LI-RADS categories, and clinical data consistently outperform single-modality approaches for survival prediction. But two challenges stand in the way. Most evidence to date comes from retrospective, single-institution studies, creating a pressing need for large-scale, multicenter prospective validation across diverse etiologies and treatment settings. And an implementation gap persists: radiology reports still focus predominantly on diagnosis and tumor size, rarely mentioning risk features in a standardized way. The authors propose that future LI-RADS iterations formally incorporate prognostic notations, flagging high-risk imaging features in report templates much as synoptic pathology reports do, so that the information reliably reaches multidisciplinary tumor boards. If that integration succeeds, the humble radiology report could become what it has quietly been all along: a noninvasive molecular portrait of the tumor, guiding each patient’s therapy to match the biology of their disease.

Subject of Research: Evolution of the LI-RADS liver imaging system from a diagnostic tool into a prognostic biomarker for hepatocellular carcinoma

Article Title: Beyond diagnosis: evolving LI-RADS from a diagnostic tool to a prognostic biomarker in hepatocellular carcinoma

Article References: Lee, J. M., & Wei, X. (2026). Beyond diagnosis: evolving LI-RADS from a diagnostic tool to a prognostic biomarker in hepatocellular carcinoma. Holistic Integrative Oncology, 5(1), Article 38. https://doi.org/10.1007/s44178-026-00245-0

Image Credits: AI Generated

DOI: 10.1007/s44178-026-00245-0

Keywords: LI-RADS, hepatocellular carcinoma, liver cancer, prognostic biomarker, radiology, tumor imaging, microvascular invasion, immune checkpoint inhibitors, treatment response, artificial intelligence, radiomics, precision oncology