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

In a pilot study published in Holistic Integrative Oncology, researchers from Liaoning Cancer Hospital and Dalian University of Technology evaluated a sprayable fluorescent probe that targets gamma-glutamyl transpeptidase, an enzyme abbreviated γ-GGT or GGT2. The enzyme sits on the membranes of mammalian cells and is normally abundant in secretory organs such as the liver, kidney, and pancreas, but its expression and activity rise sharply in many malignancies, including liver, lung, colorectal, breast, and ovarian cancers. Previous work has suggested that γ-GGT helps tumors progress, invade, and resist chemotherapy by reshaping intracellular redox metabolism, and elevated pretreatment serum levels of the enzyme have been linked to advanced stage and poor prognosis in epithelial ovarian cancer. That combination of high tumor expression and clinical relevance made γ-GGT an attractive molecular bullseye for intraoperative imaging.

The probe itself, designated NIR-SN-GGT, is a product of a medical-engineering collaboration between the cancer hospital and the State Key Laboratory of Fine Chemicals in Dalian. Chemically, it couples a dicyanoisophorone-based near-infrared fluorophore to a γ-glutamyl recognition unit through an amide bond. In its intact state the molecule is essentially dark: the γ-glutamyl group acts as an electron-withdrawing quencher that suppresses intramolecular charge transfer, the process responsible for fluorescence. When the probe encounters γ-GGT on the surface of a cancer cell, the enzyme cleaves the γ-glutamyl bond, releasing a free amine that restores charge transfer and triggers a rapid turn-on fluorescence signal in the near-infrared region, with an emission maximum around 650 nanometers, excitation near 445 nanometers, and a large Stokes shift of roughly 205 nanometers. Because the reaction occurs only where the enzyme is abundant, background fluorescence in healthy tissue stays negligible, which is precisely what a surgeon needs to distinguish tumor from normal tissue in real time.

Crucially, the team did not stop at cell cultures or mouse models, the point at which most targeted fluorescent probes have historically stalled. Instead, they built a three-tier validation pipeline that moved from bioinformatics to organoids, then to frozen tissue sections, and finally to fresh human specimens removed during surgery. Analysis of The Cancer Genome Atlas confirmed elevated GGT2 mRNA in ovarian tumors compared with normal ovary, and immunohistochemistry on archival paraffin blocks from 36 high-grade serous ovarian cancer cases showed robust γ-GGT protein on cancer cell membranes. The researchers then grew patient-derived ovarian cancer organoids in three-dimensional matrix cultures and observed distinct probe fluorescence within these miniature tumors, confirming that the probe can label γ-GGT-positive cells in a system that mimics the tumor microenvironment far better than flat cell monolayers.

The next tier involved frozen sections cut at 100-micrometer thickness from freshly resected cancer and pericancerous tissue. After spraying the sections with a 10 micromolar probe solution and waiting just one minute, confocal microscopy revealed a fluorescence pattern that colocalized almost perfectly with DAB immunostaining for γ-GGT, with markedly stronger signal in cancerous regions than in adjacent non-cancerous tissue. This spatial agreement between enzyme activity imaging and classical pathology provided the bridge from in vitro validation to real human tissue, setting the stage for the ex vivo experiments that form the heart of the study.

Between February and November 2023, the team collected fresh tissue from 16 patients aged 46 to 77 who underwent primary, interval, or secondary cytoreductive surgery for epithelial ovarian cancer, most of them at advanced FIGO stage III or IV. From these patients they harvested 42 confirmed cancerous lesions, including 18 primary ovarian lesions, 15 omental metastases, and 9 peritoneal metastases, spanning diameters of 0.5 centimeters, 0.3 centimeters, and smaller than 0.3 centimeters. Each specimen was sprayed with probe solution at 1, 5, or 10 micromolar within 30 to 40 minutes of resection, then imaged every minute for half an hour and again at one hour and five hours using a simple setup: an iPhone 12 mini camera fitted with an emission filter, working at a distance of 15 to 20 centimeters under dark conditions. Signal-to-background ratios, calculated as average lesion fluorescence divided by average fluorescence of surrounding normal tissue, served as the primary metric, with values above 1.2 considered visible to the naked eye.

The optimization experiments delivered clear answers. Across all 42 lesions, every sample produced a signal-to-background ratio above 1, yielding 100 percent fluorescence detection, and 95.2 percent were visible without any image processing. The 10 micromolar concentration proved to be the minimum effective dose, achieving signal-to-background ratios of 1.40 plus or minus 0.17 for primary lesions and 1.32 plus or minus 0.12 for omental metastases, significantly better than the 1.17 and 1.27 values seen at lower concentrations. Just as importantly, the imaging window proved remarkably forgiving: fluorescence appeared within one minute of spraying and remained stable from one minute to one hour, with ratios hovering between 1.26 and 1.29, before fading by five hours. Lesion size made no difference, with sub-0.3-centimeter implants imaged just as brightly as larger ones, a finding with obvious implications for catching the millimeter-scale seedings that surgeons most often miss.

The clinical validation phase then put the optimized protocol to the test on 27 sub-0.3-centimeter lesions from 8 patients, along with adjacent normal tissues and a negative control of normal peritoneum. Fluorescence imaging identified 26 of the 27 cancerous lesions, a true-positive rate of 96.3 percent, and 23 lesions were visible to the naked eye, a macroscopic detection rate of 81.5 percent. Only one false positive appeared, in a sample of normal peritoneal tissue that pathology confirmed lacked γ-GGT expression; the spurious signal was attributed to residual red blood cells on the tissue surface, whose protoporphyrin content autofluoresces in the same spectral region. Immunohistochemistry confirmed membrane γ-GGT expression in 100 percent of cancerous lesions in both cohorts while normal tissues stained negative, underscoring the target’s specificity. The authors note that meticulous intraoperative hemostasis and thorough saline rinsing of tissue surfaces before spraying are essential precautions to keep blood-related false positives in check.

Statistical analysis of the clinical variables added a note of caution for future use. Univariate analysis found that preoperative serum γ-GGT levels and treatment modality both influenced imaging quality, with patients whose serum levels reached at least 20 units per liter showing higher signal-to-background ratios. In multivariate analysis, only the preoperative treatment approach remained independently predictive, most likely because neoadjuvant chemotherapy can reduce γ-GGT expression or activity in tumor tissue and thereby weaken probe binding. Encouragingly, clinical stage, lesion location, and pathological type did not alter probe performance, suggesting the technique is robust across the diverse clinical presentations of ovarian cancer.

The authors are candid about the study’s limits. The cohort of 16 patients is too small for firm statistical conclusions, and larger trials will be needed to stratify effects of specific chemotherapy regimens and timing of interval surgery. The single false negative occurred in a γ-GGT-positive lesion buried more than three millimeters beneath the tissue surface, a reminder that near-infrared light at 650 nanometers penetrates tissue only shallowly; deeper lesions may require surgical exposure or next-generation probes emitting in the second near-infrared window above 1000 nanometers. Standardized fluorescence evaluation criteria will also be needed to reduce observer variability. Even so, the head-to-head comparison is striking: the FDA-approved folate receptor-alpha probe pafolacianine detects ovarian cancer lesions with 83.0 percent sensitivity and a 32.7 percent false-positive rate, whereas the sprayed γ-GGT probe achieved a 98.6 percent true-positive rate with only a 1.4 percent false-positive rate in the ex vivo experiments, while topical delivery slashes the dose required compared with intravenous agents and reduces the risk of systemic toxicity. If larger clinical trials confirm these results, a bottle of enzyme-activated dye and a filtered smartphone camera could one day become standard equipment in the fight against one of medicine’s most elusive cancers.

Subject of Research: Sprayable γ-GGT-activated fluorescent probe for intraoperative visualization of epithelial ovarian cancer

Article Title: Exploration of the application of the sprayed γ-GGT fluorescent probe for visual imaging of epithelial ovarian cancer: a pilot study

Article References: Liu, Z., Zhang, M., Li, H., Liu, L., Chen, F., Li, M., Gao, Y., Wang, D., & Yang, Z. (2026). Exploration of the application of the sprayed γ-GGT fluorescent probe for visual imaging of epithelial ovarian cancer: a pilot study. Holistic Integrative Oncology, 5(1), Article 46. https://doi.org/10.1007/s44178-026-00268-7

Image Credits: AI Generated

DOI: 10.1007/s44178-026-00268-7

Keywords: ovarian cancer, fluorescent probe, gamma-glutamyl transpeptidase, fluorescence imaging, intraoperative imaging, cytoreductive surgery, organoids, signal-to-background ratio, near-infrared fluorescence, tumor imaging, pafolacianine, precision oncology