Introduction
Intrahepatic cholangiocarcinoma (iCCA) ranks as the second most common primary malignant liver tumor, originating proximal to the second-order bile ducts of the liver,1 and its incidence continues to rise.2 Although recent advances in laparoscopic cholecystectomy have reduced the mortality rate of extrahepatic cholangiocarcinoma, the mortality rate of iCCA remains high.3 In the early stages, iCCA is often asymptomatic, hindering early detection and resulting in many patients missing the window for surgical intervention at diagnosis. Surgical resection remains the only treatment offering the potential for long-term survival in iCCA patients. The median survival time for those who undergo surgical resection is significantly better than for non-surgical patients.4 However, iCCA exhibits limited responsiveness to chemotherapy and radiotherapy, leading to restricted therapeutic outcomes.
Recent research has focused on understanding the pathological characteristics and underlying mechanisms of iCCA. It has been revealed that iCCA harbors unique genetic mutations and distinct molecular subtypes, which are instrumental in guiding treatment selection and prognostic assessment.5 Targeted therapy and immunotherapy offer promising potential, yet their effectiveness remains suboptimal, in part due to the considerable heterogeneity of iCCA. Thus, beyond achieving early and accurate diagnosis to enable surgical intervention, it is imperative to explore the tumor’s heterogeneity in depth. This includes elucidating its molecular variations and subtypes to identify more precise therapeutic targets. Medical imaging plays a crucial role in clinical decision-making and treatment planning. Leveraging imaging to reflect histological and molecular characteristics represents a critical advancement toward precision medicine, offering a foundation for more effective targeted therapies. This review will summarize recent progress in the histological classification, genetic alterations, molecular subtypes, and associated imaging signatures of iCCA. The integration of imaging with molecular and histopathological data marks a new era in iCCA management, facilitating visualization of the tumor’s pathological basis and advancing personalized therapeutic strategies.
Histological profile of iCCA
Based on growth patterns, iCCA is classified into three types: mass-forming (MF), periductal infiltrating, and intraductal growing iCCA, with MF-iCCA being the most common.6 iCCA is further categorized into large duct type (LD-iCCA) and small duct type (SD-iCCA) based on tumor location, structural characteristics, tumor cell morphology, and mucin production (Fig. 1).7
LD-iCCA predominantly occurs near the perihilar region of the liver and is characterized by columnar-like cells forming large glandular structures. It typically shows moderate to poor differentiation and exhibits abundant mucin production.8,9 Immunohistochemistry reveals positivity for S100P, MUC5AC, MUC6, and TFF1, along with positive staining for mucin in LD-iCCA.10 Serum levels of carcinoembryonic antigen and carbohydrate antigen 19-9 are often elevated in patients with LD-iCCA. LD-iCCA demonstrates a higher incidence of lymph node metastasis, vascular invasion, perineural invasion, intrahepatic dissemination, and recurrence compared to SD-iCCA.9,10 Consequently, overall and disease-free survival rates for LD-iCCA are worse than those for SD-iCCA.7,9,11
SD-iCCA is primarily located in the peripheral regions of the liver and is predominantly composed of MF-iCCA. It consists of densely arranged cuboidal-like cells forming small glandular structures, with moderate to well differentiation and an absence of mucin-secreting cells.8 Typically, the central area of the tumor displays prominent fibrous stroma. Immunohistochemistry shows positivity for C-reactive protein, N-cadherin, and CD56 in SD-iCCA.12 Song et al.13 utilized single-cell sequencing to discover and validate that S100P and SPP1 can effectively distinguish between LD-iCCA and SD-iCCA, suggesting their potential as biomarkers.
Imaging signatures of the histological subtypes of iCCA
The histological subtypes of iCCA, namely LD-iCCA and SD-iCCA, exhibit substantial prognostic differences, with LD-iCCA generally associated with a poorer prognosis. Additionally, these subtypes harbor distinct genetic mutations, suggesting different targets for effective therapies. Accurately predicting iCCA subtypes using medical imaging holds significant clinical importance.
Nam et al.14 conducted a study involving 82 pathologically confirmed cases of iCCA—22 cases of LD-iCCA and 60 cases of SD-iCCA—assessed using preoperative contrast-enhanced computed tomography (CT). Their statistical analysis revealed that arterial hyperenhancement, round or lobulated contours, and absence of bile duct encasement were more commonly observed in SD-iCCA (Fig. 2). The combination of these three CT features yielded a specificity of 90.9% and a sensitivity of 56.7% for predicting SD-iCCA. Notably, arterial hyperenhancement was identified as the most effective feature for distinguishing SD-iCCA from LD-iCCA, which typically exhibits hypoenhancement on CT (Fig. 3). These enhancement patterns are closely related to the pathological basis of the subtypes: SD-iCCA often presents with densely packed tumor cells surrounding central fibrous stroma, while LD-iCCA typically exhibits a diffuse arrangement of fibroblastic and collagen matrix.8 Park et al.15 classified SD-iCCA and LD-iCCA based on magnetic resonance imaging (MRI) features in a study of 140 pathologically confirmed MF-iCCA cases, including 93 SD-iCCA and 47 LD-iCCA. They identified four key imaging features—an infiltrative contour, diffuse biliary dilatation, absence of arterial phase hyperenhancement, and vascular invasion—that were more frequently associated with LD-iCCA (Fig. 4). The presence of two or more of these features provided a specificity of 95.7% and a sensitivity of 59.6% for distinguishing between the subtypes. The arterial enhancement pattern was particularly important for differentiation, consistent with findings from Nam et al.14 The two studies based on contrast-enhanced CT for differentiating SD-iCCA and LD-iCCA suggest that imaging-based differentiation of these subtypes is feasible with high specificity, although sensitivity remains relatively low.
This limitation in sensitivity was addressed by Xiao et al.,16 who analyzed preoperative MRI features of 63 cases of SD-iCCA and 30 cases of LD-iCCA, all pathologically confirmed. Using univariable and multivariable logistic regression analysis, they identified key imaging features that differentiate the subtypes. Independent predictors for LD-iCCA included hypoenhancement in the arterial phase (AP), intrahepatic duct dilation, and absence of targetoid restriction (Fig. 5). A predictive model incorporating these features achieved an area under the curve (AUC) of 0.91, with a sensitivity of 80.0% and specificity of 93.7%. This model demonstrated high specificity and improved sensitivity, enhancing the overall reliability of imaging-based iCCA subtype prediction. These findings reinforce the notion that arterial enhancement patterns reliably reflect the underlying pathological differences between the subtypes.
SD-iCCA and LD-iCCA also exhibit distinct characteristics on 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET)/CT. Kozaka et al.17 investigated 14 cases of iCCA (seven SD-iCCA and seven LD-iCCA) and observed that the maximum standard uptake value (SUVmax) could effectively distinguish between the subtypes, with SD-iCCA showing significantly lower SUVmax than LD-iCCA. Additionally, patients with LD-iCCA had significantly lower overall survival compared to those with SD-iCCA. Although the study sample was small, the observed differences suggest clear biological and imaging disparities between the subtypes.
A summary of the imaging features predictive of iCCA histological subtypes is presented in Table 1.14–17 While CT, MRI, and 18F-FDG PET/CT modalities reveal distinct differences between SD-iCCA and LD-iCCA, it is important to acknowledge the limitations due to small sample sizes and single-center study designs. Future research should aim to include larger cohorts and multi-center studies to enhance the reliability and generalizability of this classification system.
Table 1Imaging signatures predicting histological subtypes in intrahepatic cholangiocarcinoma
| Reference | Modality | Sample size | Imaging signatures predicting LD/SD-iCCA | Performance of predictive model |
|---|
| Nam et al.14 | CT | 82 | SD-iCCA: arterial hyperenhancement, round or lobulated contour, absence of bile duct encasement | Specificity: 90.9%; Sensitivity: 56.7% |
| Park et al.15 | MRI | 140 | LD-iCCA: infiltrative contour, diffuse biliary dilatation, no arterial phase hyperenhancement, vascular invasion | Specificity: 95.7%; Sensitivity: 59.6% |
| Xiao et al.16 | MRI | 93 | LD-iCCA: hypoenhancement in AP, intrahepatic duct dilation, lack of targetoid restriction | AUC: 0.91; Specificity: 93.7%; Sensitivity: 80.0% |
| Kozaka et al.17 | PET/CT | 14 | LD-iCCA: SUVmax | N/A |
Genetic alterations in iCCA
In recent years, there has been a growing body of research dedicated to genetic mutations associated with iCCA. This research has significantly enhanced our understanding of the mechanisms underlying the development and progression of iCCA and contributed to the refinement of its molecular subtypes, thereby facilitating targeted therapy.
Common genetic alterations in iCCA include mutations in IDH1/2, KRAS, TP53, ARID1A, CDKN2A, PTEN, FGFR2, and HER2. KRAS, TP53, and CDKN2A mutations are frequently observed in LD-iCCA, whereas IDH1/2 mutations and FGFR2 gene fusions and rearrangements are typically exclusive to SD-iCCA (Fig. 1).18 Mutations in KRAS, TP53, and CDKN2A are associated with shorter overall survival,19 suggesting their role in the poorer prognosis of LD-iCCA compared to SD-iCCA. The pattern of genetic mutations in iCCA may also correlate with hepatitis B virus (HBV) infection. Patients positive for hepatitis B surface antigen exhibit a higher frequency of TP53 mutations, whereas KRAS mutations are almost exclusively found in hepatitis B surface antigen-negative iCCA patients.20 A previous study demonstrated that the spectrum of genetic mutations differs between iCCA patients with chronic advanced liver disease and those with normal liver function,21 indicating distinct pathogenic mechanisms in these two conditions.
iCCA is characterized by substantial heterogeneity, even more so than hepatocellular carcinoma (HCC). The IDH mutation subgroup of iCCA exhibits greater tumor heterogeneity than the non-IDH mutation subgroup, with reduced T-cell infiltration and lower T-cell cytotoxicity. This suggests that IDH mutation status could aid in optimizing the selection of patients for immunotherapy.22 In patients with IDH1 mutations, those treated with the IDH1 inhibitor ivosidenib show improved overall survival compared to the placebo group, thereby reinforcing the potential of targeted therapy in iCCA.23
Imaging and radiomics features of genetic alterations and expressions in iCCA
Tumor heterogeneity is primarily reflected in differences in genetic mutations and expression profiles among tumor cells, which result in diverse biological behaviors. However, acquiring such information typically requires pathological tissue, and biopsies often represent only a partial view of the tumor, introducing selection bias and failing to capture the full tumor landscape. Non-invasively and comprehensively predicting genetic mutations and expression profiles through medical imaging could significantly lower the barrier for molecular targeted therapy by aiding in the identification of appropriate patients.
Heterogeneous tumor enhancement on imaging may reflect vascular abnormalities and hypoperfusion resulting from a hypoxic microenvironment. To investigate the relationship between imaging phenotypes of iCCA and hypoxia-related molecular features, Sadot et al.24 conducted a correlational analysis of qualitative CT imaging features, texture characteristics, and hypoxia biomarkers (e.g., EGFR, VEGF, CD24). They found that qualitative imaging features such as the “tumor-liver CT attenuation difference” and “attenuation heterogeneity within the tumor” were associated with VEGF expression, while bile duct dilation correlated with CD24 expression. Specific texture features based on the gray-level co-occurrence matrix also showed significant associations with EGFR and VEGF expression.
IDH mutations are commonly observed in iCCA, and the targeted therapy ivosidenib has shown good safety and efficacy.25 Aherne et al.26 attempted to correlate CT imaging features of iCCA with genetic information (IDH1, chromatin, and RAS-MAPK pathway-related genes), but found no significant associations. However, Zhu et al.,27 using a broader range of dynamic contrast-enhanced CT features, discovered that contrast-enhanced CT could effectively distinguish IDH1/2 mutation status. The IDH-mutant group exhibited significantly higher frequencies of intratumoral artery, rim-and-internal enhancement in AP compared to the wild-type IDH1/2 group. Moreover, the mutant group had higher CT values and enhancement ratios in both AP and portal venous phases (PVP), with the CT value in the PVP showing the highest discriminatory power (AUC = 0.798). Zhu et al.28 manually delineated tumor margins across axial CT slices in all contrast phases (precontrast, AP, PVP, and equilibrium phase (EP)) to extract texture features. Using support vector machine algorithms, five predictive models were constructed to differentiate IDH mutation status. Among them, the PVP-based classifier performed best, achieving an accuracy of 86.3%, AUC of 0.813, sensitivity of 72.7%, and specificity of 88.5% in the validation group.
Gu et al.29 found that high expression of the leucine-rich repeat kinase 2 (LRRK2) gene was associated with worse prognosis in iCCA. They also suggested potential links between LRRK2 expression and pathways related to tumor proliferation and metastasis. Notably, the ratio of the maximum to mean CT value and the subtraction of the mean from the maximum CT value in the EP significantly differed between high- and low-expression groups. The ratio metric yielded the best performance for distinguishing LRRK2 expression levels, achieving an AUC of 0.752.
Most radiogenomic studies to date have utilized CT imaging. However, 18F-FDG PET also holds promise for predicting genetic mutations. Patients with KRAS-mutated iCCA tend to have poorer prognoses, and their tumor cells exhibit increased glucose uptake and glycolytic activity. Ikeno et al.30 investigated the correlation between 18F-FDG uptake in iCCA and KRAS mutations and found that the KRAS-mutated group had significantly higher metabolic tumor volume and total lesion glycolysis than the wild-type KRAS group. Elevated metabolic tumor volume was associated with KRAS mutations and poor postoperative outcomes. In another study, Ahn et al.31 found that the SUVmax of iCCA tumors, as measured by 18F-FDG PET, was significantly correlated with genes involved in glycolysis, gluconeogenesis, phosphorylation, and cell cycle regulation. Pathway analysis revealed enrichment of pathways related to cell cycle, cell division, hypoxia, inflammation, and metabolism in patients with high SUVmax. Haghighat Jahromi et al.32 demonstrated that SUVmax from 18F-FDG PET/CT correlated with tumor mutational burden, a key biomarker of immunotherapy responsiveness. Thus, SUVmax may play a role in stratifying iCCA patients for treatment selection.
The above-mentioned radiogenomic studies are summarized in Table 2.24,26–31 Most were conducted approximately four to five years ago and are largely retrospective. Although radiogenomics is still in its early stages, it holds significant promise for the future of precision medicine. Moving forward, radiologists must work closely with clinical departments in multidisciplinary teams to conduct prospective studies using larger datasets for reproduction and validation, thereby enhancing prediction reliability. The integration of spatial transcriptomics, single-cell sequencing, radiomics, deep learning, and other multi-omics approaches may herald a new era in radiogenomics.
Table 2Summary of radiogenomic studies on intrahepatic cholangiocarcinoma
| Reference | Modality | Sample size | Gene-related imaging signatures/predictive model | Performance of predictive model |
|---|
| Sadot et al.24 | CT | 25 | expression of VEGF: “tumor-liver CT attenuation difference”, “attenuation heterogeneity within the tumor”, certain texture features based on the GLCM expression of CD24: bile duct dilation expression of EGFR: certain texture features based on the GLCM | N/A |
| Aherne et al.26 | CT | 66 | IDH mutation: significant CT features are not found | N/A |
| Zhu et al.27 | CT | 88 | IDH1/2 mutation: predictive model based on CT value of tumors in the PVP | AUC: 0.798 |
| Zhu et al.28 | CT | 138 | IDH1/2 mutation: PVP classifier built by SVM based on texture features | AUC: 0.813; Accuracy: 86.3%; Sensitivity: 72.7%; Specificity:88.5% |
| Gu et al.29 | CT | 57 | expression of LRRK2: the ratio of max to mean CT value in the EP | AUC: 0.752 |
| Ikeno et al.30 | 18F-FDG PET/CT | 50 | KRAS mutation: MTV, TLG | N/A |
| Ahn et al.31 | 18F-FDG PET/CT | 22 | Genes related to glycolysis and gluconeogenesis, phosphorylation and cell cycle: SUVmax | N/A |
Molecular subtypes of iCCA
In addition to classifying iCCA into LD-iCCA and SD-iCCA, significant progress has been made in identifying molecular subtypes of iCCA, driven by advancements in single-omics and multi-omics research.
Origin-specific molecular subtypes of iCCA
The cellular origin of iCCA remains unresolved; however, molecular subtyping focused on origin-specific traits may provide better insight. Oishi et al.33 identified two subgroups of iCCA through transcriptomic analysis based on mRNA and microRNA expression patterns: 1) HpSC-ICC, which shares gene expression profiles similar to HCC with hepatic stem cell-like gene expression traits; 2) MH-ICC, which shows gene expression features of mature hepatocytes. These findings suggest a potential common cellular origin between iCCA and HCC. The study also revealed that the altered miR-200c–EMT pathway may be crucial for maintaining stem-like features in iCCA cells, which are associated with poor prognosis. Targeting this unique miR-200c–EMT gene axis may offer a promising therapeutic strategy to treat iCCA and improve patients’ outcomes. Similarly, Jeon’s study34 highlighted an intrinsic connection between the origins of HCC and iCCA. RNA sequencing identified four subtypes (LC1–LC4), shedding light on the continuous molecular spectrum between the two. LC1 is a typical HCC subtype characterized by active bile acid metabolism and telomerase reverse transcriptase promoter mutations. LC2 is an iCCA-like HCC, marked by progenitor cell-like traits and TP53 mutations. LC3 is an HCC-like iCCA, predominantly presenting as SD-iCCA (LC3-SD), often associated with HBV infection, and shows significant expression of genes related to metabolism, inflammation, and immunity. LC4 can be further subdivided into LC4-SD and LC4-LD based on pathological features. LC4-SD differs from LC3-SD in its fibrous stroma composition: LC3-SD contains predominantly mature and intermediate stroma, whereas LC4-SD is composed mainly of immature and intermediate stroma, exhibiting more frequent chromosomal instability and immune exhaustion, and is associated with worse prognosis. LC4-LD expresses cell cycle- and proliferation-related genes, features predominantly immature stroma, and has the poorest prognosis among iCCA subtypes. This study provides valuable insights into the molecular continuum between HCC and iCCA, laying the foundation for further understanding the origins and progression of primary liver cancer.
Inflammation-associated molecular subtypes of iCCA
Different researchers have independently identified inflammation-related subtypes in iCCA, but the characteristics and prognoses of these subtypes vary across studies. Sia et al.35 classified iCCA into two main subclasses based on gene expression profiles, high-density single nucleotide polymorphism arrays, and mutation analysis. The first subclass, termed the inflammation class, is primarily marked by the activation of inflammatory signaling pathways, overexpression of cytokines, and STAT3 activation. The second subclass, the proliferation class, is characterized by the activation of oncogenic signaling pathways (including RAS, MAPK, and MET), DNA amplification at 11q13.2, deletions at 14q22.1, and mutations in KRAS and BRAF. Patients in the proliferation class generally have a poorer prognosis. Sorafenib, a kinase inhibitor that targets multiple protein kinases such as VEGFR, PDGFR, and RAF kinases, may be effective in treating proliferation-class iCCA. Conversely, JAK-STAT inhibitors are likely to elicit a better therapeutic response in inflammation-class iCCA. Bao et al.36 employed single-cell transcriptomic sequencing combined with genomics and proteomics to classify iCCA into three molecular subtypes: chromatin remodeling subtype, metabolism subtype, and chronic inflammation subtype. The metabolism subtype is associated with low-level inflammatory activity, while APOE+C1QB+ macrophages play a crucial role in reconstructing the chronic inflammation subtype, which is linked to a poorer prognosis. Given these findings, APOE+C1QB+ macrophages may represent a potential therapeutic target for patients with the chronic inflammation subtype. In animal experiments, the use of a CSF1R-blocking antibody to target and deplete APOE+C1QB+ macrophages has been shown to reduce tumor burden and cell proliferation. Dong et al.37 conducted proteogenomic analysis and identified four iCCA subtypes: S1 (inflammatory subtype), S2 (mesenchymal subtype), S3 (metabolic subtype), and S4 (differentiated subtype). The S1 subtype is characterized by abundant expression of inflammatory proteins (such as CD14, MPO, and C5AR1), KRAS mutations, elevated carbohydrate antigen 19-9 levels, a higher proportion of tumor necrosis, and intrahepatic metastasis, resulting in the shortest median survival time. The S2 subtype exhibits high levels of proteins associated with cancer-associated fibroblasts and the extracellular matrix, including FAP, POSTN, and FLT1, and is associated with a higher rate of lymph node metastasis. The S3 subtype is marked by elevated levels of MAPK and metabolism-related proteins, a predominance of TP53 mutations, and high HBV positivity. The S4 subtype maintains the highest expression levels of adhesion and biliary-specific proteins (including ANXA4, KRT18, and EPCAM). Patients in the S4 group, who have the longest median survival time, frequently display FGFR2 alterations as well as BAP1 and IDH1/2 mutations. These subgroups (S1–S4) not only demonstrate specificity in protein expression, genetic alterations, and prognosis but also exhibit distinct characteristics in terms of microenvironmental dysregulation, tumor microbiome composition, and potential therapeutic targets.
Immune landscape-defined molecular subtypes of iCCA
The immune microenvironment in tumors has garnered significant attention, with various studies focusing on its role in the classification of iCCA. Job et al.38 employed consensus clustering to categorize iCCA, highlighting differences in immune cell expression patterns and immune functions within the tumor microenvironment. They identified four subtypes: subtype I1 (immune desert pattern), subtype I2 (reactive immunogenic pattern), subtype I3 (myeloid-rich pattern), and subtype I4 (mesenchymal-rich pattern). Subtype I1 is marked by minimal expression of tumor microenvironment signatures and a paucity of immune cells. Subtype I2 is immune-active, enriched with both innate and adaptive immune cells, exhibiting significant activation of inflammatory and immune checkpoint pathways, along with abundant activated fibroblasts and quiescent hepatic stellate cells. Patients with subtype I2 have the longest median survival. Given the low immune cell presence, subtype I1 is likely to be insensitive to immunotherapy. In contrast, subtype I2, characterized by an abundance of immune cells, may derive greater benefit from immunotherapy. Subtype I3 is defined by myeloid cell enrichment, with moderate to strong expression of monocyte-derived and myeloid signatures, slightly lower expression of fibroblast signatures, and low expression of lymphoid signatures. Subtype I4 exhibits mesenchymal characteristics with robust expression of activated fibroblast signatures and elevated expression of tumorigenic factors involved in transforming growth factor-beta and integrin signaling, extracellular matrix remodeling, epithelial-mesenchymal transition, and angiogenesis—correlating with the worst prognosis. Integrating genomics, transcriptomics, proteomics, and immunomics, Lin et al.39 determined three immune subgroups of iCCA: IG1 (immune-suppressive), IG2 (immune-exclusion), and IG3 (immune-activated). These subgroups present distinct clinical, genetic, and molecular characteristics. The IG3 subtype may respond favorably to immune checkpoint inhibitor therapy, whereas the IG1 and IG2 subtypes may not be suitable candidates. Furthermore, the immune subgroups exhibited significant differences in both overall survival and recurrence-free survival, with IG1 associated with the poorest prognosis and IG3 with the most favorable outcomes. In another study, Lin et al.40 combined genomics, transcriptomics, and immunomics to classify iCCA into three subgroups based on tumor heterogeneity and immune infiltration: 1) Sparsely infiltrated subgroup, characterized by the loss of active clonal neoantigen copy number; 2) Heterogeneously immune-infiltrated subgroup, which plays a significant role in subclonal evolution across different tumor subregions; 3) Highly infiltrated subgroup, marked by extensive immune activation and a similar T-cell receptor repertoire across tumor subregions. The highly infiltrated subgroup may respond well to immunotherapy, while the sparsely infiltrated subgroup is less likely to benefit. For the heterogeneously immune-infiltrated subgroup, combination therapies may be necessary due to the high heterogeneity in immune cell infiltration. The researchers also developed a robust classification system to stratify iCCA patients into high- and low-immune evasion groups, with the low-immune evasion group exhibiting a more favorable prognosis. Chen et al.41 employed whole-exome sequencing, bulk and single-cell RNA sequencing, methylation microarray analysis, and multiplex immunostaining to analyze tumor and immune heterogeneity in multifocal iCCA. They categorized iCCA into high and low intertumor heterogeneity groups, observing consistent immune profiles across both. Notably, they found consistent responses to anti-PD-1 immunotherapy across multiple tumors. Unsupervised clustering of immune markers identified low- and high-immune subtypes. The high-immune subtype showed increased immune cell infiltration, closer tumor-immune cell interactions, and upregulated IFN-signature expression. In contrast, the low-immune subtype exhibited reduced responses to anti-PD-1 therapy and poorer recurrence-free and overall survival. Collectively, these studies suggest that immune-rich subtypes generally have more favorable prognoses.
The various molecular subtyping methodologies discussed above exhibit both overlapping features and unique strengths. Future studies with larger cohorts and more comprehensive approaches are necessary to integrate these classifications and identify patient subgroups most likely to benefit from specific therapies. Such integration will facilitate patient stratification, guide treatment selection, and enable personalized precision therapies.
Imaging and radiomics features of molecular subtypes in iCCA
The determination of molecular subtypes in iCCA currently relies on tumor tissue samples. If molecular subtypes could be predicted non-invasively through imaging, it would not only spare patients the discomfort and risks associated with tumor biopsy but also reduce costs and the consumption of medical resources. Additionally, such predictive capabilities would significantly enhance the implementation of personalized precision treatment based on molecular subtypes.
Jeon et al.34 identified four molecular subtypes—LC1 through LC4—using consensus clustering analysis of RNA transcription profiles in HCC and iCCA. LC1 typically presents as isointense or hyperintense in the hepatobiliary phase of gadoxetic acid-enhanced MRI. Gadoxetic acid utilizes the same transporter as bile acid, reflecting active bile acid metabolism in LC1. In contrast, LC2 more frequently exhibits rim arterial phase hyperenhancement on MRI, a feature often associated with the expression of stemness genes and a macrotrabecular pattern in HCC, indicating a poorer prognosis. Moreover, rim arterial phase hyperenhancement in LC2 may correlate with TP53 mutations. Although LC1 and LC2 are primarily composed of HCC, they still exhibit imaging differences that correspond to distinct molecular subtypes at the molecular level.
Immune checkpoint inhibitors exert their therapeutic effects through the activation of CD8+ T cells. iCCA can be categorized into inflammatory and non-inflammatory immunophenotypes based on CD8+ T-cell density. Preoperative identification of these immunophenotypes is crucial for predicting the potential efficacy of immune checkpoint inhibitors. Zhang et al.42 extracted texture features from multiphase and multi-sequence MRI images of iCCA. They found significant correlations between iCCA immunophenotypes and three wavelet-based radiomic features (wavelet-HHH_GLSZM_SizeZoneNonUniformityNormalized, wavelet-HLH_firstorder_Median, and wavelet-HLL_GLSZM_SizeZoneNonUniformityNormalized), as well as one 3D shape feature (original_shape_Flatness). By combining these features, they developed a model with an AUC of 0.919, demonstrating excellent performance in distinguishing immunophenotypes and identifying individuals who may benefit from immunotherapy.
Discussion and outlook
iCCA is characterized by high malignancy and heterogeneity and is typically associated with a poor prognosis. Mutations in the TP53 gene are linked to worse outcomes in iCCA and are more frequently observed in LD-iCCA.18,19 Previous studies by Zou et al.20 and Dong et al.37 identified a positive correlation between TP53 mutations and HBV infection. However, does HBV infection itself correlate with poorer outcomes in iCCA? Interestingly, a meta-analysis indicated that iCCA patients with HBV infection tend to have a more favorable prognosis compared to those without HBV infection.43 Zou et al.20 further observed that among HBV-positive individuals, those with TP53 mutations experienced worse outcomes than those without such mutations. Additionally, Huang et al.44 confirmed that TP53 and RAS/RAF mutations serve as independent prognostic factors for HBV-related and non-HBV-related iCCA, respectively. These findings underscore the need for further investigation into the relationships between liver disease etiologies (e.g., HBV infection), genetic mutations, and prognosis in iCCA. In terms of imaging, Jeon et al.34 proposed a potential association between TP53 mutations and rim arterial phase hyperenhancement, which warrants further exploration.
Targeted therapy and immunotherapy are promising yet underexplored treatment strategies for iCCA. Studies on the molecular subtypes of iCCA have shown that subtypes characterized by active immune cell infiltration are associated with better prognoses.38–41 These findings suggest that the development of therapies aimed at promoting immune cell recruitment to tumor sites may be a viable strategy. As research on molecular features and subtypes of iCCA progresses, the complexities of tumor heterogeneity are gradually being unraveled. Based on current findings, there is potential for conducting comparative studies on treatments tailored to specific genetic alterations and subtypes, thereby refining stratification approaches and identifying iCCA populations likely to benefit from targeted or immune therapies.
Medical imaging holds significant potential for the preoperative prediction of molecular characteristics and subtypes in iCCA. We observed that researchers predominantly utilize CT imaging over MRI for predicting gene expression and mutations in iCCA. Compared to MRI, CT provides superior spatial resolution and more easily interpretable parameters—such as CT values and enhancement ratios. Additionally, CT scans are more commonly performed in clinical practice, making image acquisition more accessible. These factors likely contributed to CT being the preferred modality for early radiogenomic studies in iCCA. Furthermore, considering the high costs associated with genetic sequencing, CT serves as a practical and effective alternative for predicting gene expression and mutations. Notably, substantial research has utilized MRI for gene expression and mutations prediction in other tumors, such as gliomas.45,46 Therefore, it is expected that MRI-based radiogenomics in iCCA will gain increasing attention in future studies.
To better elucidate the valuable imaging features incorporated into predictive histological studies of iCCA mentioned in the literature, we collected data from our institution’s medical record review system using the following inclusion and exclusion criteria. Inclusion criteria were: a single hepatic mass, pathologically confirmed as either SD-iCCA or LD-iCCA, and imaging performed within two months prior to surgery. Patients with a history of antitumor therapy were excluded. Images from cases with the highest number of positive predictive features were selected for illustration (Fig. 2–5). Among the 18 LD-iCCA cases at our institution, only eight (44.44%) were correctly classified as LD-iCCA using the classification method proposed by Park et al.15 This highlights the presence of single-center sample bias. While classification models may perform optimally within single-center cohorts, caution is necessary when generalizing their application. This underscores the need for multi-center studies to reduce sample bias. Nonetheless, the use of non-invasive imaging methods to predict molecular and histological subtypes remains a vital area for future research. Such approaches could support precise treatment selection and further the advancement of individualized precision medicine in iCCA management.
Conclusions
Currently, molecular subtyping of iCCA remains largely within the domain of basic research, with a significant gap in clinical application. If subtypes could be predicted—or if high-risk patients with genetic mutations could be identified—through imaging, clinicians would be better equipped to make informed diagnostic and therapeutic decisions. This represents an ideal scenario, and we hope that future research will continue to progress in this direction.
Declarations
Acknowledgement
Special thanks to Yanci Zhao, Lixia Zhang and Meng Wang for their assistance in image collection, and to Yanyan Zhu and Junhan Pan for their valuable suggestions during the conceptualization and layout of the review.
Ethical statement
All clinical images included in this review were de-identified and obtained from institutional archives with approval from the institutional review board. The requirement for individual informed consent was waived due to the retrospective and non-interventional nature of this review.
Funding
This study was funded by the Zhejiang Provincial Natural Science Foundation Committee–Zhejiang Society for Mathematical Medicine Joint Fund Major Project (LSD19H180003) and the National Natural Science Foundation of China (12090020 and 12090025).
Conflict of interest
The authors have no conflict of interests related to this publication.
Authors’ contributions
Conceptualization, image collection and evaluation (HH, FC), literature search, data analysis, writing-original draft (HH), writing-review & editing, and funding acquisition (FC). All authors have approved the final version and the publication of the manuscript.