Gastroenteropancreatic neuroendocrine tumors in 2025: From molecular profiling to artificial intelligence-driven therapy

Abstract

Gastroenteropancreatic neuroendocrine tumors represent a biologically complex and clinically heterogeneous group of neoplasms with a steadily increasing global incidence. Advances in molecular profiling have identified distinct genetic landscapes across tumor subtypes, with alterations in multiple endocrine neoplasia type 1, death domain-associated protein, X-linked mental retardation and alpha-thalassemia syndrome protein, tuberous sclerosis complex 2, and phosphatase and tensin homolog commonly defining well-differentiated neuroendocrine tumors, while tumor protein P53, retinoblastoma 1, Kirsten rat sarcoma viral oncogene homolog, and B-Raf proto-oncogene mutations are characteristic of poorly differentiated neuroendocrine carcinomas. Although tumor differentiation remains a fundamental clinical framework, it does not fully capture the extensive intertumoral and intratumoral heterogeneity that drives variable clinical behavior and therapeutic response. In this context, the integration of multi-omics data with radiomics and artificial intelligence is reshaping diagnostic and prognostic paradigms, enabling more objective assessment of proliferative indices such as Ki67 and improving risk stratification. Advances in functional imaging, including ⁶⁸Ga-DOTATATE positron emission tomography/computed tomography, together with emerging biomarkers such as circulating tumor DNA and the NETest, are further enhancing the precision of disease monitoring. Moreover, digital pathology and machine learning approaches show promise in overcoming sampling bias and interobserver variability. This narrative review synthesizes recent insights into molecular pathogenesis, diagnostic innovations, and artificial intelligence-driven therapeutic perspectives, highlighting the transition toward an integrative, data-driven model of gastroenteropancreatic neuroendocrine tumor management that bridges biological complexity with personalized clinical decision-making and optimized patient outcomes.

Key Words: Gastroenteropancreatic neuroendocrine tumors; Precision medicine; Molecular profiling; Artificial intelligence; Theranostic imaging; Liquid biopsy biomarkers; Targeted therapies

Core Tip: Gastroenteropancreatic neuroendocrine tumors exhibit marked biological and clinical heterogeneity that extends beyond traditional histological classification. While molecular features distinguishing well-differentiated neuroendocrine tumors from poorly differentiated carcinomas provide a critical biological framework, they represent only one dimension of disease complexity. This review highlights how the integration of molecular profiling, artificial intelligence–based analytics, radiomics, liquid biopsy biomarkers, including NETest 2.0, and theranostic imaging enables a more comprehensive and dynamic approach to diagnosis, risk stratification, and therapeutic decision-making. We propose an integrative precision medicine model that addresses tumor heterogeneity and supports individualized, data-driven management strategies across the full spectrum of gastroenteropancreatic neuroendocrine tumors.

INTRODUCTION

Gastroenteropancreatic neuroendocrine tumors (GEP-NETs) constitute a biologically and clinically heterogeneous group of neoplasms characterized by variable disease behavior and outcomes. Growing recognition of their complex biology has driven the development of innovative diagnostic and therapeutic strategies, including advances in molecular profiling, theranostic imaging, and the emerging application of artificial intelligence (AI) to refine clinical decision-making[1-3]. Traditionally, GEP-NETs have been classified according to primary site, functional status, and tumor grade, with the distinction between well-differentiated neuroendocrine tumors (NETs; G1-G3) and poorly differentiated neuroendocrine carcinomas (NECs) based on morphology and Ki67 proliferation index serving as a foundational framework[4-9]. Diagnostic evaluation integrates histopathology with immunohistochemical detection of neuroendocrine markers such as synaptophysin and chromogranins, complemented by functional imaging targeting somatostatin receptor (SSTR) expression, for which updated recommendations have recently been provided, particularly in nonfunctioning pancreatic NETs (PNETs)[5-9]. Current therapeutic strategies, including surgery, somatostatin analog-based biotherapy, peptide receptor radionuclide therapy (PRRT), and targeted agents, especially in PNETs harboring driver mutations such as multiple endocrine neoplasia type 1 (MEN1), death domain-associated protein (DAXX), and X-linked mental retardation and alpha-thalassemia syndrome protein (ATRX), form the cornerstone of contemporary GEP-NET management[7-10].

Within this clinical framework, the distinction between well-differentiated NETs and poorly differentiated NECs remains a critical biological and prognostic paradigm, as it directly informs therapeutic selection and expected clinical outcomes[5,6,8]. However, this binary classification alone does not fully encompass the extensive heterogeneity observed across GEP-NETs. Substantial variability exists within well-differentiated NETs themselves, driven by factors such as primary tumor site, functional status, and diverse molecular alterations[7,9,10]. Increasing evidence demonstrates pronounced intertumoral and intratumoral heterogeneity, including discordant molecular profiles across metastatic lesions within individual patients, which contributes to heterogeneous treatment responses and therapeutic resistance[1-3]. Consequently, while tumor differentiation provides an essential starting point, a broader integrative approach is required to capture the biological complexity that underlies clinical variability in GEP-NETs.

Despite significant progress, major unmet clinical needs persist. Reliable tools for predicting treatment response remain limited, and heterogeneous outcomes are frequently observed even within molecularly defined subgroups, underscoring the translational gap between molecular discoveries and their clinical implementation[1,3,4]. The lack of validated biomarkers capable of definitively distinguishing high-grade (G3) NETs from biologically distinct NECs continues to pose a major clinical challenge, particularly in therapeutically ambiguous cases[5]. In this context, this narrative review aims to synthesize contemporary evidence on molecular profiling, critically examine current diagnostic and therapeutic paradigms, and explore how advanced computational methodologies, particularly AI and machine learning (ML), can integrate multi-dimensional data to address tumor heterogeneity across the GEP-NET spectrum[1,3,4]. Using tumor differentiation as a paradigm rather than an exclusive focus, we highlight emerging strategies that support a transition toward an AI-enhanced, molecularly informed, and individualized model of care consistent with the principles of precision medicine[11-14].

LITERATURE SEARCH

This narrative review was designed to synthesize current evidence on GEP-NETs, with particular emphasis on the biological and clinical distinction between well-differentiated NETs including NET G3 and poorly differentiated NECs, as well as the heterogeneity among different NET subtypes. The scope of the review focused on three main thematic areas: (1) Molecular profiling and genetic heterogeneity underlying tumor differentiation; (2) AI-driven diagnostic, prognostic, and predictive approaches; and (3) Advances in theranostic imaging and targeted therapeutic strategies.

A comprehensive literature search was conducted in PubMed, Scopus, Web of Science, and Google Scholar for publications from 2015 to 2025. The search strategy combined Medical Subject Headings terms and free-text keywords using Boolean operators, including “gastroenteropancreatic neuroendocrine tumor” OR “GEP-NET” AND “molecular profiling” OR “precision medicine” OR “artificial intelligence” OR “radiomics” OR “theranostic imaging”. No restrictions regarding language were applied.

Eligible studies included original research articles, clinical trials, translational studies, and high-quality review articles addressing molecular mechanisms, diagnostic innovations, prognostic tools, or therapeutic advances related to GEP-NETs. Studies focusing on non-GEP-NET neoplasms, pediatric-only populations, isolated case reports without molecular data, and non–peer-reviewed publications were excluded. After title, abstract, and full-text assessment, a total of 116 articles were selected for qualitative synthesis and thematically organized according to tumor differentiation, molecular alterations, diagnostic strategies, and clinical implications.

EPIDEMIOLOGY

GEP-NETs represent a remarkably heterogeneous group of malignancies. While historically considered rare, their diagnosed incidence has increased globally[15-18]. This apparent rise is primarily attributed to heightened clinical awareness and significant advancements in diagnostic capabilities, including wider accessibility and improved resolution of cross-sectional imaging and endoscopic techniques, rather than a true surge in tumorigenesis[17,19,20]. GEP-NETs constitute the most prevalent subgroup of neuroendocrine neoplasms, accounting for > 60% of all NET cases[21]. The frequent delay in diagnosis, with > 50% of patients presenting with regional or distant metastases at diagnosis, represent a crucial concern[17,22]. This proportion increases to approximately 80% among individuals with high-grade neoplasms[21].

GEP-NET epidemiology is variable across regions. North America and Europe exhibit consistently higher incidence rates than Asian countries[16]. For example, age-adjusted incidence rates per 100000 population reach 6.22 in Norway and 5.45 in the United States, contrasting with lower rates in Japan (3.53) and China (0.8)[16]. Furthermore, primary tumor sites differ geographically: Small intestinal NETs (SiNETs) are most prevalent in Western countries, whereas rectal and PNETs predominate in Japan and China, respectively[16]. In Europe, the most frequent sites of origin are the small intestine and pancreas[17].

The median age at diagnosis for GEP-NETs is consistently in the early 60 years[17,23]. However, some studies have suggested that patients with metastatic disease tend to be diagnosed at younger ages[18]. Large registry-based analyses frequently indicate a slight female predominance (approximately 52%)[16,24]. Conversely, specific subtypes, including gastrinomas, demonstrate a higher incidence in males, with male-to-female ratios of 1.5-2:1[16,23,25]. Although site-specific sex differences exist, female patients generally demonstrate superior overall survival (OS) outcomes[16]. In the multi-ethnic population of the United States, African-American individuals demonstrate the highest incidence rates for gastric, small intestinal, and rectal NETs[16,26].

The anatomical origin of GEP-NETs influences their clinical course and incidence. For example, gastrinomas, which are rare functional tumors, have an annual incidence of only 1-1.5 cases per million[25]. Most gastrinomas arise from either the duodenum (70%) or pancreas (25%)[23]. Notably, the increased utilization of endoscopy has contributed to a dramatic increase in gastric and rectal NET detection in recent decades[24]. Delayed diagnosis constitutes a common challenge. For gastrinomas, the average time from symptom onset to diagnosis can exceed 5 years, and approximately 25% of patients present with metastatic disease at initial diagnosis[27]. Prognosis significantly differs based on the primary site, even in the metastatic setting. Based on 5-year OS rates for metastatic disease, tumors arising from the appendix (median OS, > 30 years) and rectum (24.6 years) exhibit the most favorable prognosis. In contrast, SiNETs have an intermediate prognosis (5-year OS, 69%), whereas pancreatic (median OS, 3.6 years; 5-year OS, 50%) and gastric/colorectal (5-year OS, 30%) primary tumor sites are associated with poorer outcomes[22]. GEP-NET diagnosis and treatment are evolving, with ongoing research aiming to improve patient outcomes[27]. Although AI and ML hold promise in diagnosis and management[18,28], further research is warranted to refine their clinical application and understanding of molecular characteristics[29].

MOLECULAR PATHOGENESIS AND ETIOLOGY

GEP-NETs have a multifactorial etiology, characterized by a wide spectrum of genetically and biologically heterogeneous tumors, with a fundamental distinction lying in the molecular divergence between well-differentiated NETs and poorly differentiated NECs[29-31]. These neoplasms frequently arise through the following two principal mechanisms: As components of inherited cancer predisposition syndromes or, more frequently, as sporadic occurrences[31,32]. Despite this etiologic diversity, hereditary and sporadic GEP-NETs converge on a limited number of oncogenic, epigenetic, and microenvironmental pathways that drive tumor initiation, progression, and therapeutic vulnerability across the disease spectrum (Figure 1).

Figure 1 Key pathways driving gastroenteropancreatic neuroendocrine tumor development and progression. VEGFR: Vascular endothelial growth factor receptor; IGF-1R: Insulin-like growth factor receptor 1; SSTR: Somatostatin receptor; RAS: Rat sarcoma viral oncogene homolog; BRAF: B-Raf proto-oncogene; MAPK: Mitogen-activated protein kinase; MENIN: Multiple endocrine neoplasia protein; ERK: Extracellular-signal-regulated kinase; EWSR1: EWS RNA binding protein 1; PI3K: Phosphatidylinositol-3-kinase; AKT: Protein kinase B; TSC1/2: Tuberous sclerosis 1 and 2; mTOR: Mammalian target of rapamycin; 4-EBP1: Eukaryotic translation initiation factor 4E-binding protein 1; DAXX: Death domain-associated protein; ATRX: X-linked mental retardation and alpha-thalassemia syndrome protein; ATM: Ataxia telangiectasia; MGMT: O6-methylguanine-DNA methyltransferase; PLC: Phospholipase C; SHIP1: Src homology region 2 domain-containing phosphatase-1; HIF1: Hypoxia-inducible factor 1; VHL: Von Hippel-Lindau; miR: Micro RNA; SK6: Ribosomal protein S6 kinase; PTEN: Phosphatase and tensin homolog deleted.

Hereditary syndromes account for approximately 5%-10% of cases[29,32,33]; however, both hereditary and sporadic forms ultimately converge on shared molecular mechanisms that disrupt normal cellular regulation. The recent 2022 World Health Organization Classification of Endocrine and NETs further consolidates nomenclature, highlighting the distinction between well-differentiated NETs (graded G1, G2, and G3 based on proliferation) and poorly differentiated NECs (inherently high-grade), with the Ki67 index being a significant classification tool[34].

Molecular drivers of well-differentiated NETs

A subset of GEP-NETs arises in the context of inherited tumor predisposition syndromes involving germline alterations in tumor suppressor genes that are frequently mirrored by somatic mutations in sporadic disease[31,33]. Among these, MEN1 represents the most prevalent hereditary cause, particularly in PNETs. MEN1 mutations at chromosome 11q13 result in loss of menin, a key regulator of chromatin architecture and transcriptional control via the mixed lineage leukemia histone methyltransferase complex, thereby promoting epigenetic instability and tumorigenesis[31,33,35]. Beyond hereditary disease, somatic MEN1 mutations constitute the most frequent genetic alteration in sporadic PNETs, present in approximately 35%-44% of cases[31,33,35].

Additional hereditary syndromes further illustrate the molecular diversity of NET pathogenesis. Von Hippel-Lindau (VHL) syndrome, driven by inactivation of the VHL tumor suppressor, leads to stabilization of hypoxia-inducible factors (HIFs) and consequent overexpression of pro-angiogenic mediators such as vascular endothelial growth factor (VEGF), contributing to the hypervascular phenotype characteristic of many NETs[29,31,33]. Neurofibromatosis type 1 predisposes to NET development through loss of neurofibromin, resulting in constitutive activation of the rat sarcoma viral oncogene homolog (RAS)/mitogen-activated protein kinase and phosphatidylinositol-3-kinase (PI3K)/protein kinase B (AKT)/mammalian target of rapamycin (mTOR) signaling cascades[29,31,33]. Similarly, tuberous sclerosis complex, caused by tuberous sclerosis complex 1 (TSC1) or TSC2 mutations, promotes unchecked mTOR signaling and aberrant proliferation[31,33].

In sporadic NETs, the genomic landscape varies substantially according to the primary site of origin[35,36]. In PNETs, alterations in DAXX and ATRX, genes involved in chromatin remodeling and telomere maintenance, occur in a combined 43% of cases and are typically mutually exclusive[29]. Loss of DAXX or ATRX expression is strongly associated with chromosomal instability and activation of the alternative lengthening of telomeres (ALT) pathway, a telomerase-independent mechanism conferring replicative immortality[29,33]. Concurrently, mutations affecting the PI3K/AKT/mTOR axis, including phosphatase and tensin homolog deleted (PTEN), TSC2, and PIK3CA, are observed in approximately 14% of PNETs, often independently of MEN1 alterations, underscoring pathway-level convergence despite genetic heterogeneity[31,34,36].

In contrast, SiNETs exhibit a markedly lower somatic mutational burden[31,35]. Their tumorigenesis appears to be predominantly driven by large-scale chromosomal alterations, most notably loss of chromosome 18, alongside epigenetic dysregulation, including widespread DNA hypomethylation[29,35]. Cyclin-dependent kinase inhibitor 1B, encoding cyclin-dependent kinase inhibitor p27, represents the most frequently mutated gene in SiNETs, although it is detected in only 8%-9% of cases, highlighting the dominant role of non-point mutation mechanisms in these tumors[29,31,35].

Molecular pathogenesis of poorly differentiated NECs and high-grade NETs

Poorly differentiated GEP-NECs constitute a biologically distinct entity, characterized by aggressive clinical behavior and a molecular profile that diverges from that of well-differentiated NETs[29,31]. NECs harbor alterations in canonical tumor suppressor genes, particularly tumor protein P53 (TP53) and retinoblastoma 1 (RB1), with TP53 mutations reported in 64%-95% of cases and RB1 mutations present in more than 60% of cases[30-33,37,38]. The combined loss of p53 and RB protein function represents a defining molecular feature that distinguishes NECs from NETs, which instead preferentially exhibit alterations in MEN1, DAXX, or ATRX[34].

Additional recurrent mutations in NECs include adenomatous polyposis coli, Kirsten RAS, and B-Raf proto-oncogene, with mutation frequencies varying according to the anatomical site of origin. Clinicopathological series and integrative reviews have highlighted the marked heterogeneity of gastroenteropancreatic neuroendocrine neoplasms, underscoring that high-grade digestive NECs cannot be considered a biologically uniform group[39]. Importantly, the molecular resemblance between GEP-NECs and small-cell lung cancer is incomplete, challenging the historical extrapolation of small-cell lung cancer-based treatment paradigms to all high-grade digestive NECs[39].

The World Health Organization 2019 classification formally recognized a distinct subgroup of high-grade, well-differentiated tumors designated NET G3, defined by preserved neuroendocrine architecture with a Ki67 index exceeding 20%[32,34]. Molecular profiling has confirmed that NET G3 tumors are genetically aligned with lower-grade NETs rather than NECs, frequently harboring mutations in MEN1 (21%), ATRX (17%), and DAXX (14%), while largely lacking TP53 and RB1 alterations[34]. This molecular dichotomy underscores the critical importance of accurate histopathological and molecular classification, as NET G3 and NECs represent biologically and therapeutically distinct entities despite overlapping proliferative indices[34].

Integrated pathophysiological mechanisms and tumor biology

Beyond discrete driver mutations, GEP-NET development and progression are shaped by complex interactions between tumor cells and their microenvironment. Angiogenic signaling plays a central role, with HIF-1α stabilization driving VEGF-mediated neovascularization and contributing to the characteristic tumor vascularity[29,35]. Epigenetic dysregulation further amplifies tumor heterogeneity, encompassing aberrant DNA methylation patterns, histone modifications, and the regulatory activity of noncoding RNAs, all of which modulate transcriptional programs and cellular plasticity[40].

At the functional level, well-differentiated NETs retain lineage-specific hormone secretion, reflecting their origin from specialized neuroendocrine cells. Epidemiological and clinical data further indicate that risk factors and biological behavior differ substantially between neuroendocrine neoplasms and other gastrointestinal malignancies, reinforcing the concept of tumor-specific pathophysiological trajectories[41]. Recurrent, cell-type-specific molecular alterations illustrate this coupling between genotype and phenotype. Insulinomas, glucagonomas, and somatostatinomas frequently show associations with inherited syndromes and distinct molecular backgrounds, including MEN1 and other germline alterations, underscoring lineage-dependent tumor biology[42-45]. Insulinomas often harbor somatic mutations in Yin-Yang 1, enhancing insulin gene transcription through altered promoter binding and epigenetic deregulation[46,47]. Glucagonomas demonstrate sustained activation of the glucagon receptor-cAMP-protein kinase A axis, with additional involvement of PI3K/AKT and AMP-activated protein kinase signaling pathways contributing to metabolic dysregulation and autonomous proliferation[43,48]. In contrast, nonfunctional NETs often present with advanced disease and are enriched for MEN1, DAXX, and ATRX mutations, reinforcing the link between epigenetic instability, ALT activation, and aggressive biological behavior[43,47]. Collectively, these molecular and pathophysiological mechanisms highlight GEP-NETs as a spectrum of diseases unified by neuroendocrine differentiation but with distinct genetic, epigenetic, and microenvironmental drivers. This integrated framework provides the biological foundation for contemporary classification systems and supports the transition toward molecularly informed, precision-based diagnostic and therapeutic strategies[48-50].

RISK FACTORS

Sporadic somatic mutations have been identified in DAXX, ATRX, cyclin-dependent kinase inhibitor 1B, PTEN, checkpoint kinase 2, and BRCA2, indicating defective DNA repair mechanisms, ALT pathway activation, and increased genomic instability. These alterations contribute to tumor progression and are associated with poorer prognosis. Individuals with germline MEN1 mutations experience a lifetime risk of up to 70% of developing NETs, making it the most documented hereditary link[49]. This autosomal dominant syndrome is characterized by multiple endocrine neoplasias, primarily affecting the parathyroid glands, pituitary, and pancreas. Its development comprises menin-mediated epigenetic disruption, which regulates tumor suppressor genes through the mixed lineage leukemia 1/histone H3 lysine 4 methyltransferase complex[50-53].

Other hereditary syndromes associated with NETs include neurofibromatosis type 1, VHL syndrome, and TSC1/TSC2. These conditions are inherited in an autosomal dominant manner and involve key signaling pathways, including RAS/mitogen-activated protein kinase, HIF, and mTOR[54,55]. Collectively, these syndromes account for approximately 10%-22% of NET cases, primarily presenting as nonfunctional tumors with diverse phenotypes[56,57]. Some studies have demonstrated an association between cigarette smoking and an increased risk of tumor development, particularly among heavy smokers; however, other analyses have not confirmed this association. Moreover, smoking may be linked to more advanced disease stages during diagnosis[52,57]. Conversely, moderate alcohol consumption has demonstrated an inverse correlation with the incidence of PNETs (odds ratio, 0.52; 95% confidence interval: 0.42-0.66), in contrast to its well-established role in other pancreatic malignancies[43,51].

Population-based analyses have revealed that approximately 32% of patients with sporadic PNETs reported residence in rural areas, a factor independently associated with delayed diagnosis and more advanced tumor stage at presentation. Furthermore, > 40% of patients with PNET had not completed secondary education, which is associated with lower health literacy and a decreased likelihood of undergoing screening or follow-up for nonspecific symptoms. Physical inactivity, reported in approximately 38% of cases in some cohorts in Europe, may indirectly contribute to tumorigenesis via metabolic syndrome pathways, chronic inflammation, and impaired immune surveillance. These factors, frequently coexisting with other lifestyle and environmental exposures (e.g., smoking and diabetes), underscore the significance of considering social determinants of health in both prevention strategies and healthcare access models for PNETs[49-51,55].

The chronic inflammatory environment caused by metabolic disorders, such as insulin resistance and dyslipidemia, promotes tumor formation through the long-lasting activation of the PI3K/AKT/mTOR signaling pathway and the ongoing production of reactive oxygen species[52,53]. Although these mechanisms are widely associated with cancer development, their specific role in NETs remains to be studied. Type 2 diabetes mellitus has been significantly linked to NETs, especially in cases of recent onset (< 1 year; odds ratio = 2.96; 95% confidence interval: 2.19-4.03), while chronic type 2 diabetes mellitus does not show a notable association. This indicates a possible paraneoplastic effect in the early disease stages. From a metabolic perspective, visceral obesity, dyslipidemia, and insulin resistance create a persistent inflammatory microenvironment, which encourages oncogenesis through PI3K/AKT/mTOR pathway activation and continuous oxidative stress[52,53]. However, body mass index has not been consistently shown to have a link with NETs[43,53].

CLINICAL MANIFESTATIONS

Clinical manifestations are remarkably heterogeneous and reflect the interplay between tumor differentiation, functional status, anatomical origin, and overall tumor burden. From a clinical standpoint, GEP-NETs are broadly categorized into functioning NETs (fNETs), which secrete bioactive hormones leading to characteristic endocrine syndromes, and non-fNETs (nfNETs), which lack overt hormonal hypersecretion and frequently remain clinically silent until advanced stages of disease[40,42].

Among PNETs, fNETs are most commonly encountered and present with hormone-specific symptom complexes that may precede tumor detection by months or years. Insulinomas are the most prevalent fNETs and typically manifest with recurrent hypoglycemia, characterized by neuroglycopenic symptoms, such as confusion, visual disturbances, behavioral changes, and seizures, as well as autonomic manifestations including tremor, palpitations, and diaphoresis. Weight gain due to compensatory hyperphagia is common and may obscure the underlying diagnosis, contributing to diagnostic delay[43]. Although rare, glucagonomas present with a distinctive but often late-recognized syndrome comprising progressive weight loss, diabetes mellitus, necrolytic migratory erythema, diarrhea, and stomatitis. Notably, a significant proportion of patients are diagnosed at an advanced stage, frequently with hepatic metastases at presentation[42,45,46].

Other functional PNETs tend to exhibit more subtle or nonspecific clinical manifestations. Somatostatinomas are associated with diabetes mellitus, cholelithiasis, steatorrhea, hypochlorhydria, and weight loss; however, these features are often incomplete or mild, resulting in delayed or incidental diagnosis during imaging or surgery for unrelated conditions[43,44]. Pancreatic polypeptide-secreting tumors rarely produce a distinct hormonal syndrome and typically present with vague gastrointestinal symptoms or manifestations related to tumor mass effect, such as jaundice or gastric outlet obstruction, particularly in advanced disease or in the context of MEN1 surveillance[45,49].

Functional syndromes arising from non-PNETs further contribute to the clinical complexity of GEP-NETs. Carcinoid syndrome, primarily driven by serotonin hypersecretion, classically presents with episodic facial flushing, secretory diarrhea, bronchospasm, and, in advanced cases, right-sided carcinoid heart disease due to fibrotic valvular involvement. Chronic serotonin excess may also result in niacin depletion, leading to pellagra, characterized by dermatitis, diarrhea, and neuropsychiatric disturbances[58-60]. Zollinger-Ellison syndrome, caused by gastrin-secreting tumors, manifests with severe and refractory peptic ulcer disease, gastroesophageal reflux, and chronic diarrhea resulting from pancreatic enzyme inactivation in a markedly hyperacidic environment. This syndrome is frequently associated with MEN1 and often poses significant diagnostic and therapeutic challenges[59-62].

In contrast, nfNETs often evade early clinical detection due to the absence of hormone-related symptoms. These tumors are commonly identified incidentally or present with nonspecific manifestations such as abdominal pain, jaundice, nausea, or signs of bowel obstruction secondary to local tumor growth. As a result, nfNETs are frequently diagnosed at advanced stages, with hepatic metastases or locoregional dissemination already present at the time of diagnosis[63-66]. Extrahepatic metastatic involvement, particularly to bone, may lead to localized pain or pathological fractures and is generally indicative of high tumor burden and advanced systemic disease[67-71]. Collectively, these nonspecific and late-presenting clinical features underscore the diagnostic challenges associated with nfNETs and highlight the need for integrative diagnostic approaches that extend beyond clinical assessment alone.

Given the breadth of clinical presentations and to avoid excessive descriptive redundancy, the hallmark manifestations, hormonal associations, and key diagnostic considerations of the principal functional and nonfunctional GEP-NET subtypes are summarized in Table 1. This structured overview facilitates rapid clinical reference while allowing the narrative text to emphasize diagnostic challenges and the broader implications for multimodal and precision-based diagnostic strategies.

Table 1 Clinical manifestations and diagnostic considerations of gastroenteropancreatic neuroendocrine tumors.

Tumor type/syndrome	Hormone secreted	Key clinical manifestations	Diagnostic pearls and challenges	Ref.
Insulinoma	Insulin	Fasting hypoglycemia, neuroglycopenic symptoms (confusion, seizures), autonomic symptoms (tremor, diaphoresis), weight gain	Symptoms relieved by glucose intake; tumors often small; diagnosis frequently delayed	[40,43]
Glucagonoma	Glucagon	Weight loss, diabetes mellitus, necrolytic migratory erythema, diarrhea, stomatitis	Dermatologic findings may precede diagnosis; often metastatic at presentation	[42,45,46]
Somatostatinoma	Somatostatin	Diabetes mellitus, steatorrhea, cholelithiasis, hypochlorhydria, weight loss	Subtle or incomplete syndrome; diagnosis commonly incidental	[43,44]
PPoma	Pancreatic polypeptide	Nonspecific symptoms, abdominal discomfort, weight loss; mass-effect manifestations	Lacks a distinctive hormonal syndrome; frequently associated with MEN1	[45,49]
Carcinoid syndrome	Serotonin	Episodic flushing, secretory diarrhea, bronchospasm, right-sided heart disease, pellagra	Typically requires hepatic metastases; cardiac involvement impacts prognosis	[58-60]
Zollinger-Ellison syndrome	Gastrin	Refractory peptic ulcers, severe GERD, chronic diarrhea	MEN1 association; hypergastrinemia-driven acid hypersecretion	[59-62]
Nonfunctional NETs	None	Abdominal pain, jaundice, obstruction, constitutional symptoms	Frequently advanced at diagnosis; symptoms reflect tumor burden	[40,63-65]
Bone metastases		Localized bone pain, pathological fractures	Marker of advanced systemic disease	[67-71]

NET: Neuroendocrine tumor; GERD: Gastroesophageal reflux disease; MEN1: Multiple endocrine neoplasia type 1.

DIAGNOSIS

Biochemical markers

Chromogranin A (CgA): Serum levels proportionally increase with tumor burden and are correlated with disease progression and survival. Elevated CgA levels are associated with functioning and nfNETs, particularly midgut and PNETs. Non-neoplastic elevations are observed in renal failure, chronic atrophic gastritis, and proton pump inhibitor therapy[56,61]. It is typically measured via enzyme-linked immunosorbent assay or immunoradiometric assay in serum or plasma; generally, normal values are < 95 ng/mL, although cut-off values vary slightly by laboratory. CgA remains a valuable biomarker for serial monitoring of disease burden and response to therapy[57].

Twenty-four-hour urinary 5-hydroxyindoleacetic acid: Serotonin produced by midgut NETs is metabolized to 5-hydroxyindoleacetic acid, which is excreted in the urine. Quantification is performed on 24-hour urine collections using high-performance liquid chromatography with electrochemical detection. Normal levels are < 10 mg/24 hour, and levels > 25 mg/24 hour are diagnostic of carcinoid syndrome, particularly in patients manifesting with flushing, diarrhea, and bronchospasm (Table 2)[58,59].

Table 2 Diagnostic modalities for gastroenteropancreatic neuroendocrine tumors.

Diagnostic tool	Clinical use	Limitations	Key clinical indications/strengths
Chromogranin A	General NET marker	Low specificity; elevated in benign conditions	Broadly available; useful for disease monitoring and prognosis
5-HIAA (urine)	Detects serotonin-secreting NETs	Diet-dependent; limited for non-serotonin tumors	Specific for carcinoid syndrome and metastatic midgut NETs
Hormonal assays	Confirms functional NETs (e.g., insulinoma)	Not useful in non-functional NETs	Essential for diagnosis of functioning tumors (insulinoma, gastrinoma, VIPoma)
68Ga-DOTATATE PET/CT	First-line imaging for well-differentiated NETs	Requires PET facilities; limited for SSTR-negative tumors	High sensitivity and specificity for staging, PRRT selection, and follow-up
18F-FDG PET/CT	For high-grade/aggressive NETs	Poor sensitivity in low-grade NETs	Prognostic stratification and detection of dedifferentiation
CT/MRI	Anatomic localization	Radiation (CT); MRI less available	Widely accessible; useful for surveillance and surgical planning
Endoscopic ultrasound	Detects small PanNETs, allows biopsy	Invasive, operator-dependent	Best for pancreatic and duodenal NETs < 2 cm; enables tissue acquisition
NGS panel	Identifies mutations, guides treatment	Not always actionable; cost	Enables precision therapy (e.g., mTOR-pathway, DNA-repair alterations)
CtDNA	Detects tumor DNA in blood	Still experimental; low ctDNA in indolent NETs	Non-invasive molecular profiling and disease monitoring
NETest	Monitors disease activity non-invasively	Limited availability; high cost	Early detection of recurrence and therapy response prediction

5-HIAA: 5-hydroxyindoleacetic acid; PET/CT: Positron emission tomography/computed tomography; MRI: Magnetic resonance imaging; NGS: Next generation sequencing; ctDNA: Circulating tumor DNA; SSTR: Somatostatin receptor; PRRT: Peptide receptor radionuclide therapy; mTOR: Mammalian target of rapamycin.

Specific peptide hormones

Insulin/proinsulin: During a supervised 72-hour fast, inappropriate insulin secretion is diagnostic when serum insulin is > 3 μIU/mL, proinsulin > 5 pmol/L, and C-peptide > 0.6 ng/mL, with concomitant glucose < 55 mg/dL. A proinsulin-to-insulin ratio of > 20% increases specificity for insulinoma[61].

Vasoactive intestinal peptide (VIP): Elevated in VIPomas, usually PNETs. VIP is measured using radioimmunoassay or chemiluminescence. Normal plasma levels are < 75 pg/mL. Levels > 200 pg/mL in the setting of watery diarrhea, hypokalemia, and achlorhydria (syndrome) are diagnostic[59].

Glucagon: Plasma glucagon levels > 500 pg/mL are considered suspicious, and > 1000 pg/mL is diagnostic in the appropriate clinical context. It is measured using radioimmunoassay (Table 2)[61].

Advanced imaging

⁶⁸Ga-DOTATATE positron emission tomography/computed tomography (PET/CT): It facilitates whole-body staging with near-microscopic resolution, detecting lesions as small as 5 mm. First-line use is endorsed by European Neuroendocrine Tumor Society and National Comprehensive Cancer Network guidelines for initial staging, restaging following progression, and selection for PRRT candidacy[61].

18F-aluminum fluoride-NOTA-octreotide: A novel 18F-labeled somatostatin analog with high SSTR2 affinity. As it can be cyclotron-produced, it provides higher spatial resolution, shorter acquisition time, and more convenient logistics than ⁶⁸Ga-DOTATATE in centers lacking gallium generators. Emerging data show comparable or superior diagnostic sensitivity for small liver lesions and bone metastases. This tracer is a promising option in the modern diagnostic algorithm of NETs, particularly in resource-variable settings[72].

Radiomics has evolved into a sophisticated pillar of quantitative imaging that extends far beyond basic texture analysis. By employing automated feature extraction pipelines, which leverage both hand-crafted algorithms and deep learning-based autoencoders, this advanced image analysis quantifies sub-visual tumor phenotypes and heterogeneity from standard ⁶⁸Ga-DOTATATE PET/CT and contrast-enhanced CT scans. These pipelines analyze a vast array of features, including first-order statistics, shape-based descriptors, and higher-order textural patterns (e.g., gray-level co-occurrence matrix features, entropy, and skewness). This high-dimensional data feeds into ensemble ML models, such as random forests and support vector machines, which are trained to predict histologic grade, the likelihood of metastases, and 12-month progression-free survival (PFS) with area under the curves (AUCs) often exceeding 0.80. Consequently, radiomics enables dynamically tailored surveillance intervals and provides a powerful, non-invasive biomarker for disease aggression[70-73] (Table 2).

Tissue and molecular profiling

Next-generation sequencing: Comprehensive next-generation sequencing panels detect actionable mutations, MEN1, DAXX/ATRX, TSC2, and PTEN, in pNETs. Identifying mTOR pathway aberrations in RADIANT-3 responders supports the targeted use of everolimus. TP53 or RB1 alterations signal transition to poorly differentiated NECs, requiring platinum-based chemotherapy[64,68].

Multi-omic analyses: Integrating RNA sequencing and DNA methylation distinguishes the “α-cell-like” pNET subtype, mutated ATRX/DAXX/MEN1 with pancreatic and duodenal homeobox 1 promoter hypermethylation, associated with less aggressive behavior and distinct glucagon-like peptide 1 expression[69].

NETest 2.0: A blood-based 51-gene signature scored via ML, NETest 2.0 delivers real-time assessment of tumor activity with an AUC of approximately 0.95. It forecasts radiologic progression up to 6 months before imaging changes, facilitating earlier therapeutic adjustments (Table 2)[73,74].

TREATMENT APPROACHES

Surgery: Complete surgical resection (R0) yields the best long-term outcomes in localized NETs. In pNETs ≤ 2 cm without nodal involvement, enucleation or parenchyma-sparing pancreatectomy can achieve > 90% 5-year OS. Debulking (> 70% tumor volume reduction) in metastatic disease reduces hormone-related symptoms and may prolong survival[64,69].

Somatostatin analogs: Long-acting octreotide long-acting release (LAR) (30 mg intramuscular every 4 weeks) and lanreotide autogel (120 mg subcutaneous every 4 weeks) control hormonal syndromes and slow tumor growth, with median PFS improvements from 6 to 14 months (PROMID trial)[70,71]. Dose escalation strategies guided by symptom recurrence are under investigation.

Molecular targeted therapies

Everolimus (mTOR inhibitor): RADIANT-3 demonstrated PFS extension to 11.0 months vs 4.6 months with placebo in pNETs, with manageable stomatitis and pneumonitis[72].

Sunitinib (VEGF receptor inhibitor): Improves median PFS to 11.4 months vs 5.5 months, though hypertension and fatigue require proactive management[75]. AI-driven algorithms are emerging to predict individual toxicity risk and optimal sequencing.

Radiotheranostics

¹⁷⁷Lu-DOTATATE PRRT uses a radiolabeled somatostatin analog that selectively binds to SSTR2-expressing NET cells, delivering targeted β-emitting radiation to induce DNA damage and tumor cell death. It is indicated for progressive, well-differentiated, metastatic GEP-NETs, particularly of midgut origin. In the NETTER-1 trial, PRRT significantly improved PFS to 65.2% at 20 months, compared to 10.8% with high-dose octreotide, and showed favorable safety when combined with amino acid infusions to prevent renal toxicity[76,77].

The phase III NETTER-2 trial has now extended these findings by evaluating ¹⁷⁷Lu-DOTATATE plus standard-dose octreotide LAR (30 mg) vs high-dose octreotide LAR (60 mg) as first-line therapy in patients with advanced, well-differentiated GEP-NETs of grade 2-3 (Ki67 10%-55%). NETTER-2 demonstrated a median PFS of 22.8 months for the ¹⁷⁷Lu-DOTATATE arm vs 8.5 months for high-dose octreotide alone (hazard ratio = 0.28; P < 0.0001), translating into an approximately 72% relative risk reduction of progression or death and confirming PRRT’s role as a new standard of care in this setting[78].

Next-generation alpha-emitters such as ²²⁵Ac-DOTATATE represent a further advance, delivering high-linear energy transfer α-particles with extremely short tissue penetration (< 0.1 mm). This enables highly localized cytotoxicity with minimal collateral damage, particularly valuable in PRRT-refractory disease. Early-phase studies report tumor control rates exceeding 70%, with a lower incidence of hematologic toxicity compared to β-emitters, suggesting a promising salvage strategy for aggressive NET biology[79-81].

PROGNOSIS

Tumor grade and Ki67: Ki67 index remains the cornerstone prognostic factor; grade 1 (< 3%) 5-year OS is approximately 86%, grade 2 (3%-20%) approximately 58%, and grade 3 (> 20%) approximately 25%[67,82].

Primary site: Midgut NETs (appendiceal/ileal) attain 10-year relative survival rates > 90%, whereas pancreatic and pulmonary NETs fare worse, guiding site-specific follow-up intensity[82].

PRRT response: Lack of progression at 9 months post-PRRT predicts significantly longer OS, underscoring early imaging assessment at 3-6 month intervals[64].

Metastatic burden: Presence of liver metastases reduces 5-year OS to 13%-54%, with lesion number and distribution informing eligibility for loco-regional therapies[69].

PROGNOSTIC TOOLS IN 2025

NETest 2.0: A validated liquid biopsy with approximately 95% sensitivity and approximately 81% specificity; serial NETest scores predict PFS and OS, enabling dynamic risk stratification and earlier intervention[73].

Radiomic-Ki67 models: Combining PET/CT radiomic features with Ki67 data yields an AUC of approximately 0.81 for progression prediction, refining individual surveillance schedules[63].

ML nomograms: Models trained on > 11000 Surveillance, Epidemiology, and End Results patients integrate clinical, pathologic, and imaging variables to predict distant metastasis risk and OS, outperforming American Joint Committee on Cancer/tumor node metastasis staging and aiding multidisciplinary decision making[83].

FUTURE PERSPECTIVES AND EMERGING CONCEPTS

Research in GEP-NETs is progressively converging toward integrative precision oncology models that combine molecular profiling, advanced imaging, and computational analytics. Rather than isolated innovations, AI-driven methods, patient-derived organoid platforms, radiomics, and immunogenomic strategies represent interconnected components of a unified ecosystem designed to address the profound biological and clinical heterogeneity that limits traditional classification-based management. This convergence enables dynamic disease modeling and supports individualized, biology-driven therapeutic strategies[84-86].

Within this framework, AI, primarily implemented through ML and deep learning algorithms, functions as an advanced clinical decision-support system capable of integrating multimodal data, including histopathology, radiomics, genomics, transcriptomics, and clinical variables. These approaches enhance diagnostic classification, refine risk stratification, and assist treatment planning without replacing clinical judgment. In parallel, patient-derived organoids provide functional validation platforms for molecular hypotheses, while immuno-oncology research explores biomarkers of immune responsiveness across distinct GEP-NET subsets. Collectively, these developments illustrate a paradigm shift toward adaptive, data-driven precision oncology rather than parallel technological advances[84-86].

Revolutionizing diagnostics through AI and molecular integration

One of the most transformative advances lies in overcoming the limitations of current diagnostic and grading systems, which remain vulnerable to subjectivity and sampling bias, particularly in Ki67 assessment[87,88]. The integration of digital pathology with AI-based image analysis enables automated hotspot detection and quantitative Ki67 indexing, improving reproducibility and prognostic accuracy while reducing interobserver variability[85,87,88]. ML-based survival models, including random survival forests, have demonstrated superior performance compared with traditional Cox regression by capturing non-linear interactions among clinicopathological variables such as lymph node ratio, tumor grade, and molecular alterations[85].

Concurrently, liquid biopsy technologies are emerging as minimally invasive tools for real-time disease monitoring and longitudinal risk assessment[85,89]. Circulating analytes, including mRNA, microRNA, circulating tumor DNA, and extracellular vesicles, enable dynamic evaluation of tumor activity and treatment response[85,89]. The multigene mRNA-based NETest has consistently demonstrated higher sensitivity and specificity than CgA for detecting minimal residual disease and predicting recurrence, frequently preceding radiologic progression[88,89]. While circulating tumor cells remain limited by low shedding rates in well-differentiated NETs, circulating tumor DNA has shown value in high-grade neoplasms and mixed neuroendocrine-non-neuroendocrine neoplasms by enabling molecular characterization and identification of actionable alterations[89].

The next frontier in liquid biopsy development involves AI-assisted multi-omic platforms, including nanopore sequencing technologies capable of simultaneously profiling single-nucleotide variants, structural rearrangements, and epigenetic features such as cell-free DNA methylation from a single sample[84]. These approaches hold promise for non-invasive tumor fraction quantification, detection of lineage-specific methylation signatures, and early identification of therapeutic resistance, particularly in biologically aggressive GEP-NET subsets[84,89].

Radiomics as a prognostic and theranostic platform

Radiomics is evolving from a complementary imaging tool into a central prognostic and theranostic modality. Early transcriptomic and DNA microarray studies in GEP-NETs demonstrated the existence of biologically distinct molecular subgroups with different clinical behaviors and therapeutic vulnerabilities, providing the conceptual foundation for linking imaging phenotypes with tumor biology. Longitudinal AI-driven radiomic analysis enables spatiotemporal tracking of tumor heterogeneity across serial imaging studies, facilitating early prediction of treatment resistance before conventional radiologic progression[90-92]. The concept of delta-radiomics, quantifying changes in radiomic features following a single cycle of systemic therapy, has emerged as a potential early surrogate marker of overall survival and treatment efficacy[93]. These imaging-based response dynamics are increasingly relevant in the context of systemic treatment strategies commonly used in advanced neuroendocrine neoplasms, including alkylating chemotherapy regimens[94]. In parallel, PRRT has become an established treatment modality for well-differentiated GEP-NETs, providing a robust clinical framework in which early imaging biomarkers may help refine response assessment and treatment sequencing[95].

AI-guided therapeutics and molecularly informed treatment selection

Beyond radiomics, molecularly informed treatment selection has been supported by meta-analyses demonstrating the efficacy of targeted agents across neuroendocrine tumor subtypes, reinforcing the need for individualized therapeutic strategies based on tumor biology rather than anatomical origin alone[96]. Within this context, AI-based approaches are increasingly applied to integrate imaging, molecular, and clinical data, supporting treatment selection, response prediction, and longitudinal disease monitoring in neuroendocrine tumors[97].

Next-generation radiotheranostics and combination strategies

PRRT remains a cornerstone therapy for well-differentiated metastatic GEP-NETs; however, ongoing innovations aim to enhance efficacy and overcome resistance[98,99]. Alpha-emitting radioligands such as 225Ac-labeled somatostatin analogs offer higher linear energy transfer and more potent DNA damage with reduced tissue penetration, showing promise in PRRT-refractory disease[99]. Combination strategies integrating PRRT with radiosensitizers, cytotoxic regimens (e.g., capecitabine and temozolomide), or epigenetic modulators to restore radiosensitivity are actively under investigation[88,100]. These advances support a shift toward earlier-line PRRT integration in molecularly selected patients[99].

Immunogenomics and tumor microenvironment modulation

The limited efficacy of single-agent immune checkpoint inhibitors in well-differentiated, immunologically “cold” GEP-NETs has prompted exploration of combination strategies aimed at remodeling the tumor microenvironment[87,88,92]. Approaches combining immunotherapy with anti-angiogenic agents, chemotherapy, or oncolytic viruses seek to enhance immune infiltration and antigen presentation[85,86]. Single-cell transcriptomic analyses have identified immune-enriched NET subtypes characterized by tumor-macrophage interaction pathways, providing a foundation for immunogenomic patient stratification[101-105].

The panoptic precision oncology model

Ultimately, these advances converge toward a panoptic model of precision oncology that integrates multi-omic data, radiomics, digital pathology, liquid biopsies, and clinical parameters within AI-driven analytical frameworks[84,85]. This model enables non-invasive diagnosis, dynamic disease monitoring, and objective resolution of diagnostic ambiguities, particularly in distinguishing NET G3 from poorly differentiated NEC and in deciphering intratumoral clonal heterogeneity that drives therapeutic resistance[84].

Despite its transformative potential, implementation of the panoptic model faces substantial barriers, including high infrastructure costs, lack of standardized analytical pipelines, data interoperability challenges, regulatory constraints, and concerns regarding data privacy and equity of access[85,106-108]. Addressing these challenges will require coordinated international collaboration, standardized protocols, and investment in digital health infrastructure to ensure equitable global access to AI-driven precision medicine[106-108]. Importantly, real-world experiences integrating multimodal clinical data for treatment sequencing and longitudinal disease monitoring, as well as contemporary guideline-driven frameworks for standardized care pathways in neuroendocrine tumors, provide early proof-of-concept for the feasibility of such integrative models in routine practice[109,110].

Emerging clinical tools, such as digital tumor passports, dynamic platforms integrating molecular, radiologic, and clinical data, illustrate the translational potential of this approach by facilitating multidisciplinary decision-making, personalized treatment sequencing, and adaptive disease monitoring[111-116]. Together, these innovations support a continuously adaptive model of care that aligns biological complexity with individualized therapeutic strategies, reinforcing the future of precision oncology in GEP-NETs.

CONCLUSION

In 2025, the management of GEP-NETs entered a new era driven by precision oncology and AI. Through molecular profiling, theranostic imaging, and personalized algorithms, clinicians can now stratify risk, monitor response, and optimize outcomes with unprecedented accuracy. The continued evolution of integrative AI platforms, omics-based diagnostics, and patient-centered technologies will further reshape the future of GEP-NET care.