Artificial intelligence (AI) has expanded rapidly across the oncology continuum-spanning early detection, histopathology, molecular profiling, treatment selection, drug discovery, toxicity surveillance, and survivorship care. However, these varied applications occupy fundamentally different stages of clinical and biological translation.

We conducted a critical narrative review searching PubMed/MEDLINE, Europe PMC, IEEE Xplore, arXiv, and ClinicalTrials.gov (updated to July 21, 2026). From these sources, an evidence map of 202 publications (including 182 original studies) was assembled based on study design, prospective or external validation, mechanistic rigor, clinical utility, and representation across the cancer continuum.

The primary synthesis demonstrates evidence gradient rather than uniform translation across oncology: Detection & Imaging: Select mammography and colonoscopy tools are supported by randomized controlled trials demonstrating workflow efficiency and detection gains. Conversely, negative pragmatic trials highlight that robust technical accuracy does not automatically translate into improved diagnostic pathways.

Large self-supervised and vision-language foundational models exhibit strong cross-task and cross-institutional transferability. However, prospective real-world deployment remains rare, and reporting on calibration, subgroup performance, and data provenance is inconsistent. Molecular & Systems Biology: Multi-omics, single-cell, spatial transcriptomics, graph-based, and perturbation models increasingly yield falsifiable hypotheses regarding tumor microenvironments, regulatory networks, and drug vulnerabilities; most, however, lack upstream functional validation. Therapeutic Decision Support: Applications in treatment response, surgical assistance, toxicity monitoring, and clinical large language models (LLMs) remain limited by cohort heterogeneity, temporal drift, unstandardized endpoints, and a lack of evidence that model-guided care alters patient outcomes.

AI applications must be evaluated using evidence matched directly to their specific clinical or biological claims: external validation for transportability, calibration/decision analysis for utility, prospective workflow studies for operational value, randomized trials for patient outcome benefits, and perturbation experiments for biological mechanism. This claim-matched framework clarifies the boundary between promising computational models and clinically or mechanistically credible oncology, establishing a roadmap for reliable integration into precision cancer care.

Graft-versus-host disease (GvHD) remains a major cause of morbidity and non-relapse mortality after allogeneic hematopoietic cell transplantation. Building on prior evidence implicating mitochondrial DNA (mtDNA) variation in myelodysplastic neoplasms (MDS), we investigated the prognostic impact of recipient and donor mtDNA variation on acute and chronic GvHD. Whole‑genome sequencing was performed on pretransplant samples from 494 MDS donor-recipient pairs (discovery cohort) from the Center for International Blood and Marrow Transplant Research (CIBMTR). Associations between mtDNA variations and acute GvHD grades II-IV (aGvHD24), grades III-IV (aGvHD34), and chronic GvHD (cGvHD) were assessed using cause-specific Cox proportional hazards models. Model discrimination was quantified using area under the curve (AUC). Findings were evaluated in an independent cohort of 295 MDS using targeted mtDNA deep sequencing. Among the recipient mtDNA gene associations identified in the discovery cohort, MT-ND2 for aGvHD34 (p = 1.1 × 10^-3) and MT-CYB for cGvHD (p = 3.6 × 10^-3) were independently validated, while MT-CO2, which was nominally associated with aGvHD34 in the discovery cohort, was also significant in the recipient validation cohort (p = 3.2 × 10^-3). Incorporating recipient mtDNA genes into the CIBMTR aGvHD Risk Index improved discrimination for aGvHD34 by 11% (p = 0.009). In conclusion, mtDNA variants may provide independent prognostic information for GvHD and may enhance current risk stratification when integrated with standard clinical parameters.

Glioma is an aggressive brain tumor with poor prognosis and high recurrence, which significantly affects patients' quality of life. Identifying new biomarkers and therapeutic targets is crucial for improving diagnosis and treatment. This study focuses on the SDF4 gene, which has been implicated in various cancers; however, its role in glioma remains underexplored due to limited research.

To comprehensively investigate the role of SDF4 in glioma, we integrated bulk, single-cell, and spatial transcriptomic analyses with experimental validation. This approach includes bioinformatics evaluations, RNA sequencing analysis, single-cell RNA sequencing (scRNA-seq), spatial transcriptomics, survival analyses, gene enrichment assessments, and in vitro experimental validation. Furthermore, we explored the relationship between SDF4 expression and specific clinical parameters such as tumor grade and patient prognosis. The expression levels of SDF4 in invasive glioma specimens were assessed through immunohistochemical techniques.

Our investigation reveals that SDF4 is markedly upregulated in glioma and correlates with unfavorable prognostic outcomes. Through Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression and weighted gene co-expression network analysis (WGCNA), SDF4 was identified as a pivotal prognostic biomarker; elevated levels of SDF4 were significantly linked to decreased overall survival rates. Functional enrichment analyses indicate that SDF4 regulates the cell cycle, mediates apoptosis, and remodels the extracellular matrix. Furthermore, single-cell and spatial transcriptomic studies have pinpointed SDF4 expression specifically within astrocytes, underscoring its essential function within the tumor microenvironment. In vitro experiments demonstrated that silencing SDF4 reduces glioma cell proliferation and promotes apoptosis; this effect is primarily mediated through the PI3K-Akt signaling pathway, as silencing effectively curtails its activation. Moreover, a notable elevation in SDF4 protein levels was observed in glioma patients. These elevated levels were associated with genetic mutations such as IDH1 and TERT, as well as clinical features including World Health Organization (WHO) classification and the methylation status of MGMT.

Our research underscores the critical role of SDF4 in glioma progression, highlighting its potential as a strategic target for glioma management and as a prognostic marker. Additionally, these findings pave the way for novel therapeutic strategies.

Osteosarcoma progression and treatment resistance are strongly influenced by the tumour microenvironment. One of the key players in tumour microenvironments is cancer-associated fibroblast. Much of the conventional cancer-associated fibroblast-centred research paradigm focuses on the secretory and matrix-remodelling aspects of fibroblasts. However, less is known on how fibroblasts influence osteosarcoma cell behaviour, particular under stress.

In a proof-of-concept study, human dermal fibroblasts and osteosarcoma cells were studied using direct coculture and size-restricted transwell systems. Fluorescent labelling, confocal microscopy, live-cell imaging, and single-cell tracking were used to assess intercellular transfer, migratory behaviour, proliferation, and survival under chemotherapy and hypoxic challenge.

Osteosarcoma cells acquired fibroblast-derived, mitochondria cargo-enriched material in both direct and indirect coculture systems, with transfer of material enhanced under carboplatin and hypoxic stress. Notably, fibroblast coculture increased osteosarcoma cell migration and promoted survival under stress but did not substantially increase basal proliferation.

These findings suggest that fibroblast-to-cancer cell transfer of mitochondria-enriched cargo is accelerated during stress conditions with the concomitant observation of proportional osteosarcoma protection to degree of fibroblast cargo transfer. Further research should examine stress-amplified transfer during tumour adaptation, chemoresistance, and metastatic behaviour, potentially highlighting stromal intercellular communication as an important aspect of osteosarcoma biology.

Ampullary adenocarcinoma is a rare gastrointestinal malignancy that typically occurs in older populations and characteristically presents with obstructive jaundice. Atypical biochemical presentations in young adults without hyperbilirubinemia are extremely rare and can lead to significant diagnostic delays.

A 37-year-old asymptomatic male presented with a rapid, isolated, and extreme elevation of serum gamma-glutamyl transferase (GGT), surging from 78 U/L to 1793 U/L over 6 months, and an elevated CA19-9 level. Total bilirubin, transaminases, and baseline liver parenchyma remained entirely normal. Abdominal ultrasound followed by high-resolution magnetic resonance cholangiopancreatography (MRCP) revealed extrahepatic biliary dilation and focal thickening at the major duodenal papilla. Based on high radiological suspicion, a multidisciplinary team recommended direct surgical intervention without prior endoscopic ultrasonography (EUS) or biopsy. The patient underwent a curative pancreaticoduodenectomy (Whipple procedure). Histopathology confirmed a 2.8 cm, moderately-to-poorly differentiated ampullary adenocarcinoma of pancreatobiliary origin (pT3aN0M0, R0 resection) with perineural invasion. Molecular profiling demonstrated a KRAS wild-type, BRAF wild-type, microsatellite stable (MSS), and HER2 (2+) equivocal status. Postoperatively, the serum GGT normalized completely to 57 U/L by day 18.

An isolated, rapidly rising GGT can serve as an independent early alert for occult terminal bile duct lesions before full mechanical occlusion and hyperbilirubinemia manifest. Early advanced imaging with MRCP is critical in such atypical presentations. Comprehensive molecular profiling provides additional targetable and prognostic insights for personalized postoperative surveillance.

Not applicable.

Immune checkpoint inhibitors (ICIs) have revolutionised the management of advanced malignancies, but can cause immune-related adverse events (irAEs), including vision-threatening complications such as optic neuropathy. Ocular irAEs have been described with several programmed death-1/programmed death-ligand 1 and cytotoxic T lymphocyte-associated antigen-4 inhibitors, but ocular toxicities of tislelizumab have so far been limited to Vogt-Koyanagi-Harada (VKH)-like uveitis, panuveitis and retinal vasculitis. We describe a woman with metastatic lung adenocarcinoma who was treated with second-line gemcitabine, carboplatin and tislelizumab. After the second cycle of chemo-immunotherapy, she developed painless, progressive bilateral visual blurring in the absence of intracranial disease. Ophthalmic assessment was consistent with toxic optic neuropathy in the context of ongoing tislelizumab exposure, and other causes were excluded. Tislelizumab was discontinued and high-dose intravenous methylprednisolone (1 g/day for 5 days) was administered, without ICI re-challenge. Visual function did not improve, indicating an irreversible optic nerve injury. To our knowledge, this is the first reported case of toxic optic neuropathy associated with tislelizumab, extending the spectrum of tislelizumab-related ocular irAEs beyond previously documented VKH-like panuveitis, VKH-like uveitis and retinal vasculitis. This case highlights the importance of early recognition of visual symptoms and periodic ophthalmic assessment in patients receiving ICIs, as well as decision-making regarding ICI continuation or discontinuation when sight-threatening neuro-ophthalmic irAEs are suspected.

The molecular mechanisms linking chromosome homeostasis to fibroblast reprogramming into cancer-associated fibroblasts (CAFs) remain poorly understood. Here, we identify a clinically relevant, telomere-associated nuclear pathway driving CAF activation across multiple human cancers. The long noncoding RNA TERRA is consistently elevated in CAFs from skin, lung, and breast tumors. We show that the androgen receptor (AR) represses TERRA transcription through subtelomeric binding, and its loss enables TERRA up-regulation and interaction with the RNA binding protein NONO. This interaction forms a CAF-specific nuclear complex that reprograms gene expression. Mechanistically, NONO is directly recruited to and binds the promoters of key CAF effector genes, where it regulates their transcription in a TERRA-dependent manner. Disrupting the TERRA-NONO axis, via TERRA silencing or pharmacologic NONO inhibition, reverses CAF activation and suppresses tumor-stroma interactions in vitro and in vivo, highlighting its therapeutic potential.

Myelodysplastic syndrome (MDS) is a clonal hematopoietic neoplasm with a high risk of leukemic transformation. Alternative polyadenylation (APA) is an important post-transcriptional mechanism, and its dysregulation is closely linked to the pathogenesis of hematological malignancies. This study aimed to clarify the regulatory role of the NUDT21-APA-SRSF2 axis in the progression of myelodysplastic syndrome (MDS). To this end, we established MDS cell models overexpressing or knocking down NUDT21 and SRSF2. The regulatory mechanism and functional impacts of this axis were systematically evaluated using RT-qPCR, Western blot, CCK-8 assay and flow cytometry; APA site analysis mediated by specific primers and transcriptome sequencing. The results showed that the expression of NUDT21 was negatively correlated with that of SRSF2 (p < 0.05). Overexpression of NUDT21 promoted the utilisation of distal polyadenylation sites on SRSF2, increasing the expression of its long 3'UTR variants; conversely, NUDT21 knockdown produced the opposite effect (p < 0.05). Functionally, NUDT21 enhanced MDS cell proliferation and inhibited apoptosis, though this effect was reversed by co-overexpression of SRSF2. Furthermore, SRSF2 could antagonize the regulatory effect of NUDT21 on the cell cycle. In summary, our in vitro data demonstrate that NUDT21 remodels SRSF2 isoform balance via alternative polyadenylation (APA) and regulates biological functions of MDS-derived cells. The NUDT21-APA-SRSF2 axis represents a key regulatory cascade with potential therapeutic implications.

Gallic acid (GA), a natural polyphenol, has been enormously recognized for its anticancer properties across multiple cancer types, including triple-negative breast cancer (TNBC), yet its precise molecular mechanisms underlying its effects on TNBC, particularly those related to autophagy, remain unclear. This investigation sought to elucidate the autophagy-related targets and pathways modulated by GA in TNBC by integrating network pharmacology, bioinformatics, molecular docking, and in vitro validation. Network pharmacology identified 23 overlapping targets between GA, TNBC, and autophagy-related gene sets, including key genes such as CASP3, TP53, AKT1, PIK3CA, and HIF-1 A. Protein-Protein Interaction network analysis revealed ten hub genes as central nodes in GA's action mechanism. Gene Ontology enrichment highlighted biological processes such as DNA damage response, protein phosphorylation, and regulation of apoptosis and autophagy. KEGG pathway analysis further indicated enrichment in pathways related to EGFR inhibitor resistance, insulin signaling, autophagy, and apoptosis, among others. Molecular docking demonstrated strong binding affinities between GA and the identified autophagy modulatory genes. In vitro experiments showed that GA inhibited the proliferation and migration of BT-549 and HCC1395 TNBC cells while inducing autophagy. Induction of autophagy by GA in the BT-549 and HCC1395 cell lines was proved by acridine orange, MAP1LC3B fluorescence, and western blotting of key autophagy regulators. Taken together, this integrated approach, combining network pharmacology and experimental validation, identified key targets and signaling pathways through which GA regulates autophagy in TNBC. These findings enhance our understanding of GA's anticancer mechanisms and support its plausible development as a therapeutic entity for TNBC.

​Pancreatic ductal adenocarcinoma (PDAC) remains highly lethal, with a 5‑year survival of only 13%. Its aggressive biology and late diagnosis often lead to metastatic spread at presentations. This review examines how metabolic reprogramming regulates angiogenesis in PDAC and explores the therapeutic implications of this coupling. We systematically analyzed published studies on glucose, amino acid, and lipid metabolic alterations and their interconnection with neovascularization through hypoxia-inducible factor-1α (HIF-1α), oncogenic signaling, and metabolite-mediated pathways. Based on a synthesis of the current literature, metabolic-angiogenic coupling appears to operate through at least three interconnected mechanisms: the HIF‑1α/pyruvate kinase M2 (PKM2)-driven glycolytic-vascular endothelial growth factor (VEGF) axis; lactate acting as a signaling molecule that promotes M2‑polarized tumor‑associated macrophages and stabilizes HIF‑1α; and lipid metabolites--including eicosanoids and fatty acid‑binding protein 4 (FABP4)-that modulate endothelial function and survival. In addition, glutamine-derived intermediates support endothelial sprouting and extracellular matrix remodeling, while tryptophan and serine pathways influence the immune-angiogenic balance. Importantly, compensatory pathways (e.g., HIF-1α-independent angiogenesis via glycogen accumulation and protease-activated receptor‑2 activation) limit the efficacy of single‑agent interventions, highlighting the need for combination strategies. Understanding these metabolic-angiogenic networks provides a rationale for biomarker‑guided therapies that simultaneously target tumor metabolism and vascularization. Integrating metabolic and angiogenic biomarkers (e.g., VEGF, lactate dehydrogenase A) with imaging-based metabolic profiling may improve patient stratification and treatment monitoring in PDAC.