Classic Hodgkin lymphoma (CHL) is a B-cell neoplasm differentially categorized according to the 5th Edition of the World Health Organization Classification of Haematolymphoid Tumours (WHO-HAEM5) [1] and the International Consensus Classification (ICC) [2]. The term “nodular lymphocyte predominant Hodgkin lymphoma” (NLPHL) is listed under the family of Hodgkin lymphomas in the latest edition of the WHO classification; however, a new entity, called “Nodular lymphocyte predominant B-cell lymphoma” (NLPBL), was proposed by the ICC based on major biological and clinical differences from CHL. This review focuses on CHL, which is clearly differentiated from NLPHL and NLPBL in molecular, histopathological, clinical, and therapeutic terms.

Since its first description in 1832 by Thomas Hodgkin [3] and its cytological and histopathological characterization by Carl Sternberg [4] and Dorothy Reed [5], the history of CHL constitutes a model of successful collaboration between basic and clinical research, which has led to the gradual achievement of fundamental clinical objectives [6]. Currently, the cure rate of CHL is near 90%, being a paradigm of curable cancer.

Several milestones in this translational research model in oncohematology deserve special mention. Among them are the recognition of the B-cell origin in CHL as described in two seminal studies by Küppers et al. [7,8], as well as the progressive optimization of chemotherapy [9,10,11,12,13] and radiotherapy [14,15,16] regimens throughout the twentieth century, the improvement of conditioning and supportive care [17,18,19,20,21] and management of hematopoietic stem cell transplantation (HSCT) [22,23] and the development of targeted therapies against the cluster of differentiation (CD) 30 antigen [24,25,26] and the Programmed Cell Death protein-1 (PD-1) [27] which has significantly improved the management of refractory/relapsed CHL (R/R CHL). In parallel, the de-escalation of treatment intensity based on improved assessment of therapeutic response using 18-fluorodeoxyglucose ([18F]FDG) positron emission tomography combined with computed tomography (^[18F]FDGPET/CT) [28,29,30,31], the development of more accurate [32,33,34,35] and robust [36,37] clinical prognostic models and a better comprehension of the key biological features of the disease [38,39] have directed CHL research toward its current challenges.

Despite the progress made, there are still 10%–20% of patients diagnosed with CHL who will die due to the progression of the disease. Thus, the design of strategies to enable early identification of this high-risk subset of CHL patients constitutes an essential goal in the current era. Three technologies could support the development of new prognostic models that, eventually, will capture in the future the biological complexity of high-risk CHL: liquid biopsy, metabolic evaluation through artificial intelligence (AI) assisted ^[18F]FDGPET/CT imaging, and spatially resolved analyses. The integration of these approaches into multimodal prognostic tools is expected to enable the identification of aggressive disease features both at diagnosis and during disease monitoring.

First, given that circulating tumor DNA (ctDNA) fragments are detectable in the blood of CHL patients at concentrations sufficient for extraction and sequencing (~30 ng/mL) [40], liquid biopsy will be increasingly adopted as a precise genotyping and disease-monitoring strategy. Previous studies have shown that allele frequencies found in the blood of CHL patients are higher than those detected in lymph node samples (1.99% vs. 1.6%, p = 0.02) [41]. Moreover, a longitudinal study of ctDNA concentration has shown correlation with the degree of treatment response and clonal evolution [42]. Furthermore, ctDNA concentrations are lower in patients who achieve remission compared to those who do not (p < 0.001, at a 96 months of follow-up) [43]. Liquid biopsy in CHL, therefore, allows the identification of mutations in peripheral blood that cannot be detected through tissue-based analyses and facilitates the identification of patients at risk of treatment failure, who may benefit from more intensive, dynamic, or molecular-guided therapy strategies [44,45].

Second, application of AI algorithms in PET/CT evaluation constitutes an active field of research with potential advantages in the management of analysis time, interobserver variability, and response prediction. On the one hand, evaluating PET/CT images in patients with CHL is a time-consuming and resource-intensive process. The complexity of analyzing these images can stem from the need to differentiate between tumor processes and inflammatory or infectious processes, which are common in this immunocompromised population. AI-based software for analyzing medical images using PET/CT could reduce analysis time by automating some processes and requesting a medical professional review only in particularly difficult cases. On the other hand, research areas such as interobserver variability and response prediction using AI algorithms have been examined in recent literature [46]. Particularly, in CHL, automatic segmentation and calculation of PET/CT parameters using AI tools have been compared with traditional methods. In a recent study [47], no statistically significant differences were observed for parameters used in clinical routine such as the Standardized Uptake Value maximum (SUV max) and Metabolic Tumor Volume (MTV), at either the interim or end-of-treatment assessment in a cohort of 32 CHL patients. These data suggest that the introduction of AI tools in the analysis of PET/CT medical images in patients diagnosed with CHL will be gradual in the near future, but still require standardization and normalization of the tools used.

However, novel AI algorithms are not yet standardized in the clinical setting; as a consequence, other studies have shown that AI algorithms tend to overestimate PET/CT parameters (e.g., the Deauville score) [48,49]. These “false positive” evaluations of AI methods in the analysis of PET/CT images raise the concern of unnecessary treatment escalation and higher toxicity. Thus, harmonizing the AI algorithms used for PET/CT image analysis in CHL constitutes an important clinical challenge.

Third, CHL tissue biopsies are characterized by a scarce tumoral fraction (1%–10%), termed Reed-Sternberg cells and mononuclear (Hodgkin), lacunar, and mummified variants, surrounded by a prominent tumor microenvironment (TME). The TME of CHL consists of a heterogeneous tissue space composed of T lymphocytes, B lymphocytes, macrophages, neutrophils, eosinophils, and histiocytes, among other minority cell types [50]. Traditionally, the TME of CHL has been considered the result of an impaired immune system unable to control tumor progression effectively, a perspective aligned with the “reductionist view” described by Hanahan and Weinberg in their landmark 2000 paper [51].

Nevertheless, TME of CHL constitutes a precise network of cellular, molecular, and spatially defined interactions orchestrated to enhance tumor progression and treatment resistance. The spatial localization of cellular elements, the interactions between cancerous and “non-cancerous” cell populations, and their genomic and molecular evolution in CHL are the consequence of dynamic evolutionary processes. In this regard, the discipline of eco-oncology [52] and the study of CHL from an ecological perspective can serve as a basis for a new approach to monitoring and treatment strategies for CHL. Recent evidence from digital image analysis (DIA) and computational pathology (CP) tools demonstrates that the histological architecture of CHL reflects a precise spatial topology created by cancer cells throughout its biological evolution [53,54]. The guiding role of ecological principles in cancer biology has been demonstrated by the fact that the hallmarks of cancer are spatially defined, shaping both cancer evolution and drug sensitivity [55].

In this review, we focus on the clinical and prognostic significance of spatial and topological features of TME in CHL, emphasizing the need to bring together the fields of molecular biology, DIA, CP, biomathematics, and AI computing to generate new, more accurate, and clinically relevant prognostic models in CHL. The aim of this study is to review the contributions of DIA and CP in clinical and translational research on CHL.

There is a growing body of knowledge on the applications of DIA and CP for identifying prognostic biomarkers in TME of CHL. In this milieu, pioneering studies reported interesting results, leading to a transition from traditional studies focused on CHL histopathology [56,57,58] to the novel approaches centered on slide digitization and application of DIA algorithms. In this process, scanning and feature extraction with CP algorithms constitute an opportunity to develop a real TME-based risk stratification model. This aspect constitutes a key objective in CHL research, as a fraction (10%–20%) of patients are still primarily refractory or relapse (R/R) after achieving complete metabolic response (CMR) with current chemotherapy and radiotherapy regimens [59].

In 2013, the first study linking CHL and CP was published [60]; the authors described a workflow for exploiting CD30-stained lymphatic tissue image data in CHL samples. Later, in 2016, mathematical studies by Watts and Strogatz on small-world networks [61] and by Barabási and Albert on random networks [62] enabled the construction of a cellular graph from CHL histopathological slides [63]. This cell graph of CD30⁺ tumoral (Hodgkin/Reed-Sternberg or HRS) cells is defined by two properties: 1) greater degree than a random model, i.e., the location of tumoral cells in CHL is not spatially random, but rather there is a certain degree of clustering between cells; and 2) absence of freedom of scale, i.e., there are no highly connected groups of HRS cells (defined as hubs) and a distant “disconnected” tumoral population, but rather a connection described by Watts and Strogratz as “small-world networks”. Fig. 1 (own data) shows the process of scanning, cell identification, and quantification of CD30⁺ tumoral HRS cells in CHL using the software commonly employed by our research group (MembraneQuant & NuclearQuant, QuantCenter Software, 3DHistech Ltd., Budapest, Hungary, and QuPath, version 0.4). Fig. 1 shows the cellular segmentation and quantification of CD30⁺ cells in CHL using QuPath (version 0.4).

These studies have certain limitations, such as evaluating a single cell population (CD30⁺ tumor cellularity), omitting the other cellular components that integrate the TME (T lymphocytes, B lymphocytes, plasma cells, dendritic cells, eosinophils, macrophages, among others). In 2020, Hanning et al. demonstrated the preference of tumor cells for certain neighboring cells based on their morphology [64], and recently, novel parameters such as cell velocity, number of cell contacts, and cell contact time have been obtained and evaluated in CHL preparations visualized in 3D and 4D images [65,66,67].

The aforementioned studies, however, do not examine the impact of DIA data on treatment response. Consequently, prognostic information on the evolution of CHL patients cannot be extracted. For this purpose, the first paper was published in 2021 by members of the German Hodgkin Study Group (GHSG) [68]. In this analysis, using whole slide imaging (WSI) of histopathological CHL samples from 487 patients, a low B-cell content in the TME (<21%) was identified as an adverse prognostic factor. This finding was reported in advanced-stage CHL patients who received Bleomycin, Etoposide, Adriamycin, Cyclophosphamide, Vincristine, Procarbazine, and Prednisone (BEACOPP) chemotherapy as first-line treatment.

After the GHSG research, our group reported that a percentage of 2% of HRS cells, quantified using WSI and DIA methods, statistically significantly discriminated patients achieving CMR from those R/R patients following first line treatment with Adriamycin, Bleomycin, Vinblastine and Dacarbazine (ABVD), both for overall survival (OS) (p = 0.001) and progression-free survival (PFS) (p = 0.005) [69]. Furthermore, CHL patients with a T cell percentage below 26.70% in the TME showed a statistically significantly shorter OS (p = 0.019) and PFS (p = 0.041). In addition, the combined impact of HRS and T cells percentages identified a high-risk subgroup of CHL. No prognostic impact of B cells was identified in this study, probably due to differences in cohort sizes (n = 85 vs. n = 487), distribution of disease stages (localised, intermediate, and advanced CHL vs. advanced CHL), and chemotherapy regimens (ABVD vs. BEACOPP).

Moreover, the immune response mediated by Programmed Cell Death Protein-1 (PD-1)/Programmed Cell Death Ligand-1 (PD-L1) also appears to play a prognostic role in CHL. Using the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) guidelines [70], PD-L1/CD30 ratio identified a group of patients with an increased risk of failure to ABVD chemotherapy as first-line therapy: CHL patients with PD-L1/CD30 ratio > 47.1 showed a lower OS (median OS: 53.7 months; 95% confidence interval [CI]: 28.7 to 78.7) compared to those below this threshold (median OS: 105.4 months; 95% CI: 89.6 to 121.3) (p = 0.04). When adjusted for covariates, PD-L1/CD30 ratio retained its prognostic impact for both OS (Hazard Ratio [HR]: 1.005; 95% CI: 1.002 to 1.008; p < 0.001) and PFS (HR: 3.442; 95% CI: 1.045 to 11.340; p = 0.04) based on data from three different centers.

Recently, GHSG reassessed the prognostic impact of B lymphocytes in CHL using DIA technologies [71]. On this occasion, a subgroup of patients was treated with ABVD chemotherapy (4 cycles) or ABVD and escalated BEACOPP (BEACOPPe) (defined as “2 × 2” regimen), which modified the homogeneity of the treatments included. Using a cutoff value of 8%, determined by Receiver Operating Characteristics (ROC) curves, CHL patients with ≤8% B lymphocytes in the TME showed a lower 5-year PFS (71% vs. 83%). However, these differences were not statistically significant (p = 0.08) and were not reflected in the OS data (p = 0.34).

Moreover, the combined use of WSI and next-generation sequencing (NGS) in CHL allows assessment of the combined impact of genomic and histopathological variables. In 2025, we reported [72] data on the quantification of B lymphocytes in the TME of 220 diagnostic slides from 110 CHL patients, and on the sequencing of tissue biopsies using a targeted panel that included 47 genes recurrently mutated in mature B-cell neoplasms. A combined reduction in B cell content (<8%) and the absence of mutations in apoptosis-associated genes (ABL1, BIRC3, CASP8, and FAS) were identified as adverse prognostic factors. CHL patients exhibiting this combined event (low B cells in TME and absence of mutations in apoptosis genes) presented a significantly lower OS (median OS: 31.5 months, 95% CI: 0 to 69.7 months) compared with patients without this feature (median OS: 84.7 months, 95% CI: 61.9 to 107.5 months) (p = 0.01). Additionally, this subgroup showed significantly lower PFS (median PFS: 8.5 months; 95% CI: 7.5 to 9.5 months) compared to CHL enriched in B cells or with mutations in genes linked to apoptosis (median PFS: 55.2 months; 95% CI: 42.4 to 68 months) (p < 0.001). Fig. 2 (own data) shows the process of scanning, cell identification, and quantification of CD21⁺ cells in TME of CHL using the software MembraneQuant & NuclearQuant (QuantCenter Software, 3DHistech Ltd., Budapest, Hungary, and QuPath, version 0.4). Fig. 2 illustrates the calculation of distances within the network of CD21⁺ follicular dendritic cells (FDC) in a diagnostic CHL lymph node sample using the Napari software (version 0.6.6). This spatial characterization is critical for defining the specialized niches and cellular distances that modulate the interaction between HRS cells and their surrounding microenvironment.

Currently, there are six major prognostic stratification models validated for its routine application in the clinical management of CHL patients: the model proposed by GHSG [73], the model of the European Organisation for Research and Treatment of Cancer (EORTC) [74], the National Cancer Institute of Canada (NCI-C) model [74], the National Comprehensive Cancer Network (NCCN) model [75], the International Prognostic Score (IPS), designed by Hasenclever and Diehl [32], and the Advanced-stage CHL International Prognostic Index (A-HIPI) [76], designed by the HoLISTIC Consortium (Hodgkin Lymphoma International Study for Individual Care).

The first four models (GHSG, EORTC, NCI-C, and NCCN) are used in patients with localized disease (stages I–II). In contrast, the latter two (IPS and A-HIPI) are used in patients with advanced disease, defined as stage II with B symptoms (fever, profuse night sweats, or weight loss of more than 10% in the last six months) accompanied by bulky masses or extranodal involvement, as well as stages III–IV. Table 1 describes the prognostic variables considered in each of the models currently used at the CHL for treatment strategy planning.

Risk stratification of CHL in the clinical setting constitutes a critical step, as it is required for planning the optimal therapeutic strategy (most effective and least toxic). In particular, risk categorization is needed to distinguish between patients with localized and advanced disease. Patients with localized stages should then be subclassified into favorable (I–II without risk factors) and unfavorable (I–II with risk factors or III–IV). The intensity of chemotherapy and radiotherapy schedules to be administered during the course of the treatment varies among the different subgroups.

In patients with advanced-stage disease (IIB, III–IV), incorporating IPS variables decreases 5-year OS. Specifically, CHL patients with an IPS of three or higher are classified as high risk with a 5-year OS below 59%. In addition, it should be noted that, because of its recent communication using data from >5000 CHL patients with newly diagnosed advanced-stage CHL and the widespread use expected in the coming years, the A-HIPI model has improved outcome predictions using 5-year OS and 5-year PFS as clinical endpoints compared to the traditional IPS [76].

However, none of the above models incorporate histopathological, cytogenetic, or molecular variables, as they are based exclusively on clinical, analytical, and disease extension parameters. This fact is striking in clinical research on CHL, given the progress made in understanding the molecular basis of the disease over the last 20 years. The clinical translation of preclinical results is complex, and in most cases the prognostic biomarkers that have been proposed as clinically actionable have not demonstrated external validity.

In this context, the study of the cell populations that comprise the TME is one of the most developed areas of translational research in CHL. On the one hand, the scarcity of tumor cellularity and the prominence of different TME cell populations suggest a key role for TME in the pathogenesis and natural history of the disease. Different cell types have been evaluated in recent years. Firstly, the T cell population constitutes the most abundant cell type in the CHL TME. The presence of regulatory T cells within the tumor-reactive microenvironment could explain the suppression of the antitumor immune response observed in these patients. In a previous study [77], a low infiltration of regulatory T cells (FOXP3⁺) and a high proportion of cytotoxic T lymphocytes (TIA1 and granzyme B) were associated with decreased survival. Greaves et al. [78] validated the combination of FOXP3 and CD68 as biomarkers for establishing prognostic stratification groups.

On the other hand, the work by Alonso-Álvarez et al. [79] demonstrated that in diagnostic biopsies of patients with CHL, a high proportion of CD8⁺ T lymphocytes (≥15%) and CD4⁺ T lymphocytes (≥75%) correlated with longer treatment failure-free survival (FFS) at 10 years. Of note, natural killer (NK) cell infiltration and activation in TME appear to confer a favorable prognosis in CHL [80]. Álvaro-naranjo et al. [81] reported that a low number of activated NK cells is associated with traditionally adverse prognostic factors, including the presence of B symptoms at diagnosis and the advanced disease stages. However, NK cell infiltration in the TME of CHL remains largely diminished [82], and their precise role in eliminating HRS cells remains an active field of research.

Regarding the study of the mononuclear phagocytic system (MPS) and its cellular components, the pioneering work of Steidl et al. [83] reported a negative prognostic impact for the increased CD68⁺ macrophages, both after first-line treatment (p = 0.03 for PFS; p = 0.003 for disease-specific survival, DSS) and in response to HSCT (p = 0.008), improving the predictive capacity of the IPS (p = 0.003 vs. p = 0.03). These data have been validated in subsequent studies [84], suggesting an adverse prognostic role for high levels of CD68⁺ cells in CHL.

Dendritic cells, follicular dendritic cells, and neuroendocrine cells, however, have been relatively understudied in CHL. On the one hand, an association has been reported between low levels of circulating dendritic cells [85] and survival in patients with CHL. Furthermore, employing immunohistochemistry (IHC), the reduction of CD21⁺ FDC in biopsy specimens of CHL, and the degree of destruction of the FDC network, correlated with a lower survival in CHL [86,87].

To the best of our knowledge, only one study [88] has reported aberrant expression of synaptophysin (a neuroendocrine marker) in CHL. In this study, survival analysis was not conducted. However, it is well established that the neuroendocrine axis modulates immune system activity and is expressed in other tumors as a mechanism of resistance to treatment [89,90]. Further studies are needed to evaluate the expression of neuroendocrine markers in CHL and their association with treatment responses.