The original research, conducted by Zhu and colleagues, examined dendrobine in larval zebrafish subjected to high-glucose conditions, a widely used short-term model of early diabetic retinopathy. The team combined several complementary techniques: imaging of the retinal vasculature, histological examination of retinal tissue, behavioral testing, assays of oxidative stress, and transcriptomic analysis of gene expression. Together, these approaches suggested that dendrobine attenuated a range of abnormalities induced by short-term high-glucose exposure, including the enlargement of retinal vessels that characterizes the earliest stages of the disease. At a concentration of 40 milligrams per liter, the compound appeared to reduce both vascular changes and whole-body glucose levels in the larvae, prompting the authors to propose dendrobine as a candidate therapeutic agent.

Chen’s central criticism concerns the timing of treatment, a detail that may sound technical but carries enormous clinical weight. In the original study, dendrobine and glucose were administered concurrently, from three to six days post-fertilization, meaning the compound was present in the larvae from the very moment hyperglycemic stress began. Dendrobine was never introduced after a retinal abnormality had already been established. The experiment, Chen argues, therefore answers a prevention question rather than a treatment question: it shows that dendrobine can limit injury while high-glucose stress is being induced, but it says nothing about whether the compound can reverse damage that already exists.

This distinction matters because diabetic retinopathy in human patients is almost never caught at the moment metabolic stress begins. By the time most people are diagnosed and treated, vascular and neural changes in the retina have already taken hold. A compound that reduces retinal injury when present from the onset of metabolic stress may not retain the same efficacy once those structural changes are established. Chen points out that the 130 millimolar glucose larval model used in the study was originally developed as a short-term model of early hyperglycemia-related retinal vascular change, and that subsequent work in the field has consistently distinguished such short-term larval immersion models from longer-duration diabetic models designed to study more established retinal complications. The results, in other words, support a protective effect during acute exposure but do not yet demonstrate therapeutic efficacy against established disease.

Chen also notes a second design gap: the study included no dendrobine-only group under normal glucose conditions. This omission is significant because several outcomes interpreted as rescue, including developmental and transcriptomic changes, could plausibly have been influenced by direct effects of dendrobine on normal larval development. Without knowing how the compound behaves in healthy larvae, it is difficult to attribute every observed improvement to protection against glucose injury. Chen proposes that a delayed-treatment design would address both concerns at once: researchers could first document vascular or structural retinal abnormalities after glucose exposure, then initiate dendrobine treatment while maintaining hyperglycemic conditions. Such an approach would more closely mirror the question faced in clinical practice, namely whether a drug can improve retinal injury that is already present rather than prevent it from developing.

The second major issue raised in the letter involves disentangling local retinal effects from systemic glucose lowering. Because dendrobine at 40 milligrams per liter reduced both retinal vessel enlargement and whole-body glucose levels, and because most major outcomes were assessed at the same six-day time point, it remains unclear whether the retinal benefit was a direct effect on eye tissue or simply a downstream consequence of reduced blood sugar. The original authors suggested that retinal improvement might occur before major systemic metabolic change, implying a local retinal mechanism, but as Chen observes, no serial measurements or glucose-matched comparisons were performed to establish that temporal sequence. To their credit, the authors acknowledged that systemic metabolic effects could not be separated from direct retinal actions, but Chen argues that the field needs more: retina-specific assays, direct measurements of retinal dendrobine exposure, or comparisons between experimental groups with similar systemic glucose levels would all help determine whether the compound acts on the retina independently of its glucose-lowering properties.

The third and perhaps most methodologically intricate criticism concerns the transcriptomic data. RNA sequencing in the original study was performed on whole larvae rather than isolated retinal tissue, yet the enriched biological pathways were subsequently linked to retinal protection. Chen emphasizes that whole-organism bulk transcriptomic data cannot identify which tissue or cell type is responsible for differential gene expression without additional spatial or cell-resolved information, a limitation increasingly recognized in the genomics community. Signals arising from the gut, liver, kidney, or other glucose-responsive organs could dominate the expression profile, masking or mimicking retinal-specific changes.

Compounding this concern is a statistical question. Chen notes that neither the main methods section nor the supplementary text of the original study specifies whether the reported transcriptome-wide P values were adjusted for multiple testing. This matters because false-discovery control is a standard requirement when thousands of genes are tested simultaneously; without such correction, apparent pathway enrichments can arise by chance. The quantitative reverse transcription PCR experiments in the original paper do support the direction of expression changes in selected genes, but validating a handful of genes does not establish that the implicated pathways causally mediate the retinal phenotype. At this stage, Chen concludes, the transcriptomic findings are better suited to generating candidate pathways than to identifying what the original authors called precise molecular targets. Retina-specific transcriptomic profiling and functional perturbation of the proposed pathways, such as genetic or pharmacological manipulation in larvae, would provide firmer mechanistic support.

The exchange arrives at a moment of growing interest in natural products as sources of anti-diabetic and neuroprotective compounds. Dendrobine, extracted from Dendrobium orchids long used in traditional medicine, has attracted attention for its reported anti-inflammatory and antioxidant properties, and zebrafish have become a favored platform for early-stage drug screening because their transparent larvae allow direct visualization of developing blood vessels. The high-glucose immersion model offers speed and scale that rodent models cannot match, letting researchers survey vascular changes within days rather than months. Yet the same convenience creates interpretive traps, as this letter makes clear: short exposure windows, whole-organism measurements, and concurrent treatment designs can each blur the line between protection and treatment, between a retinal effect and a systemic one.

None of this diminishes the value of the original work, and Chen is explicit on that point. The study provides useful evidence that dendrobine modifies retinal and systemic responses during acute high-glucose exposure in zebrafish larvae, and the multi-modal approach, spanning imaging, histology, behavior, oxidative stress assays, and transcriptomics, offers a template for how such screens should be conducted. What the letter demands is precision in interpretation: determining whether dendrobine can treat established retinal injury, separating local retinal effects from systemic glucose lowering, and validating the proposed pathways within retinal tissue itself. For the millions of people worldwide at risk of vision loss from diabetic retinopathy, the difference between a preventive compound and a therapeutic one is not academic. It defines the clinical trials that must eventually be run, the patients who could benefit, and the stage of disease at which any new drug must prove itself. This careful critique, published open access in the Journal of Translational Medicine, is a reminder that in translational science, the rigor of interpretation matters as much as the rigor of the experiment.

Subject of Research: Evaluation of dendrobine's protective effects in a zebrafish model of diabetic retinopathy

Article Title: Letter to the Editor regarding “Functions of dendrobine in a zebrafish model of diabetic retinopathy”

Article References: Letter to the Editor regarding “Functions of dendrobine in a zebrafish model of diabetic retinopathy”. (n.d.). https://doi.org/10.1186/s12967-026-08991-5

Image Credits: AI Generated

DOI: 10.1186/s12967-026-08991-5

Keywords: dendrobine, diabetic retinopathy, zebrafish, high glucose, retinal vasculature, transcriptomics, oxidative stress, glucose lowering, letter to the editor, translational medicine, drug screening, hyperglycemia

The original study, led by Li and colleagues, assembled an impressive array of experimental tools to implicate NIK. The team analyzed retinal transcriptomic data, exposed a human microglial cell line known as HMC3 to advanced glycation end products, the damaging sugar-derived molecules that accumulate in diabetes, and then used genetic knockdown of NIK alongside a pharmacological inhibitor called B022 to see what happened. They also set up co-culture systems in which microglia and endothelial cells, the building blocks of blood vessels, were grown together to model the crosstalk that occurs at the blood-retinal barrier. The results pointed in a consistent direction: when NIK activity was reduced, inflammatory signaling quieted down and the integrity of the barrier improved, suggesting that NIK participates in the inflammatory signaling relevant to retinal vascular injury.

But the interpretation of those results has become the flashpoint of the debate. Li and colleagues framed their findings largely through the lens of M1/M2 polarization, a long-standing model in immunology that treats macrophages and microglia as existing in two opposing states. In this binary picture, M1 cells are inflammatory warriors that pump out tumor necrosis factor-alpha, interleukin-1 beta, and interleukin-6, while M2 cells are the healers, producing anti-inflammatory mediators such as interleukin-10 and repairing tissue. According to this framework, diseases like diabetic retinopathy arise when the balance tips too far toward M1, and therapies work by pushing cells back toward M2. It is an appealingly simple story, and it has dominated the literature for nearly two decades.

The trouble, as Jing Chen of the People’s Hospital of Leshan argues in a letter to the editor, is that the study’s own data do not fit that simple story. In the AGE-treated HMC3 cells, both CD68-positive and CD206-positive cells increased, markers that under the binary model would be assigned to opposing phenotypes. At the same time, elevated levels of the inflammatory cytokines TNF-alpha, IL-1 beta, and IL-6 were accompanied by increased IL-10, the archetypal anti-inflammatory signal. Under a strict reciprocal switch model, these markers should move in opposite directions. Instead, they rose together, a pattern far more consistent with overlapping or coexisting activation programs than with cells flipping from one fixed identity to another. As Chen emphasizes, this is not a quibble over nomenclature imposed from outside; the discrepancy arises directly from the study’s own measurements.

There are also technical reasons why bulk measurements of a handful of canonical markers cannot settle the question. When a population of cells is measured in aggregate, it is impossible to determine whether apparently opposing markers are co-expressed within the same cells, whether they come from distinct cellular subsets, or whether they reflect transitions between transcriptional states. In retinal tissue, the problem is compounded because bulk measurements cannot separate resident microglia from infiltrating myeloid cells that flood in from the circulation as the barrier breaks down. And a change in a small panel of markers, however canonical, does not define a complete transcriptional state. The reduction in M1-associated markers and the increase in M2-associated markers after NIK inhibition therefore support a shift in inflammatory phenotype, but they do not by themselves establish a conversion between fixed M1 and M2 identities. What they are compatible with is something more interesting: a mixed activation landscape in which NIK influences one or more disease-associated microglial programs.

This reinterpretation is not happening in a vacuum. The microglia field has been undergoing a conceptual revolution, driven largely by single-cell RNA sequencing technologies that can profile the gene expression of thousands of individual cells at once. In 2022, a landmark consensus paper in Neuron by Paolicelli, Sierra, Stevens, and colleagues argued that the field had outgrown the M1/M2 dichotomy and should embrace a multidimensional, context-dependent view of microglial states. Under this framework, microglia do not occupy one of two poles but instead slide along continuous axes of gene expression, adopting transient, disease-specific transcriptional programs that cannot be captured by any short list of surface markers. The states observed in a dish, moreover, often bear little resemblance to the states that emerge in living tissue under disease pressure.

Single-cell studies in diabetic retinopathy specifically have made the case even more compelling. Wang and colleagues reported marked microglial heterogeneity and dynamic subtype changes in early experimental diabetic retinopathy, showing that the cellular landscape shifts as the disease progresses. Geng and colleagues, working in a rat model of non-proliferative diabetic retinopathy, identified four distinct microglial subtypes, including an SPP1-high population associated with oxidative stress, apoptosis, and injury to the inner blood-retinal barrier. And in human proliferative diabetic retinopathy, Gu and colleagues described MARCO-positive microglia with dual pro-angiogenic and pro-fibrotic properties, a phenotype that could contribute to both the abnormal vessel growth and the scarring that characterize the advanced disease. These studies differ in species, tissue source, and disease stage, and they do not define a single longitudinal trajectory, nor do they establish NIK as the driver of any of these states. But together they show why a binary framework may miss biologically relevant heterogeneity in diabetic retinopathy, and why the question raised by the NIK study needs sharper tools to answer.

The distinction matters because changes in the abundance of a particular microglial state and changes within a state are biologically different phenomena, and they could imply very different therapeutic effects. If a state-based model is correct, the goal of NIK inhibition would not simply be to suppress M1 and promote M2, but to reduce harmful NIK-dependent programs while preserving the homeostatic and reparative functions that microglia perform in the healthy retina. That is a far more nuanced therapeutic objective, and it demands far more precise experimental designs. Chen also raises a critical caveat about the original study’s in vivo work: because B022 was delivered intravitreally without microglia-specific targeting, the observed improvement in blood-retinal barrier integrity cannot establish that retinal microglial NIK is the cell-specific mediator of the effect. Other cell types in the eye, including endothelial cells and infiltrating immune cells, also rely on NF-κB pathway signaling, and the drug would have reached them all. The pharmacological benefit supports further evaluation of NIK as a therapeutic target, but it is not equivalent to microglia-specific causality.

The path forward, Chen suggests, is concrete and technically achievable. A practical first step would be to compare the retinal transcriptomic data already generated by the original study with published signatures of disease-associated microglial states in diabetic retinopathy, asking whether the genes altered by NIK inhibition overlap with known state markers. That could be followed by single-cell or spatial transcriptomic profiling of retinas after NIK inhibition, which would reveal whether the treatment changes the abundance of particular microglial subtypes, alters their transcriptional programs, or both. Even more decisive would be microglia-restricted manipulation of Map3k14, the gene encoding NIK, combined with lineage-resolved analysis that can distinguish resident microglia from infiltrating macrophages. Such experiments would directly test whether the enzyme acts within microglia to remodel pathogenic states, or whether its protective effects flow through other cellular targets entirely.

None of this diminishes the significance of the original findings; if anything, it sharpens them. The evidence that NIK participates in the inflammatory signaling that damages the blood-retinal barrier in diabetes remains intact, and the pharmacological data justify continued interest in NIK-directed therapy for diabetic retinopathy. What changes is the biological framing: rather than a dial that turns inflammatory cells from bad to good, NIK may be better understood as a regulator of pathogenic microglial state remodeling, a molecular lever that shifts the composition and behavior of a heterogeneous population of immune cells in the diabetic retina. For a disease that affects hundreds of millions of people worldwide and for which current treatments only slow, and do not reverse, the loss of vision, that refined framework could prove to be the difference between another failed anti-inflammatory strategy and a therapy that finally targets the right cells in the right state at the right time.

Subject of Research: The role of NF-κB-inducing kinase in microglial state remodeling and blood-retinal barrier dysfunction in diabetic retinopathy

Article Title: From M1/M2 polarization to disease-associated microglial states: refining the role of NIK in diabetic retinopathy

Article References: Chen, J. (2026). From M1/M2 polarization to disease-associated microglial states: refining the role of NIK in diabetic retinopathy. Journal of Translational Medicine, 24(1), Article 1190. https://doi.org/10.1186/s12967-026-08987-1

Image Credits: AI Generated

DOI: 10.1186/s12967-026-08987-1

Keywords: diabetic retinopathy, microglia, NIK, NF-κB-inducing kinase, M1/M2 polarization, disease-associated microglial states, blood-retinal barrier, single-cell transcriptomics, neuroinflammation, B022, retinal endothelial cells, MAP3K14

To understand why this matters, it helps to grasp how HIF works at the molecular level. The factor is a heterodimer composed of an oxygen-sensitive alpha subunit and a constitutive beta subunit. In well-oxygenated cells, prolyl hydroxylase enzymes tag the alpha subunit with hydroxyl groups, marking it for recognition by the von Hippel-Lindau tumor suppressor protein, which ubiquitinates HIF-alpha and condemns it to rapid destruction by the proteasome. When oxygen levels fall, this degradation machinery stalls. HIF-alpha accumulates, translocates to the nucleus, and pairs with HIF-beta to bind hypoxia response elements scattered throughout the genome. The result is a coordinated transcriptional surge: genes encoding vascular endothelial growth factor, erythropoietin, glycolytic enzymes, matrix metalloproteinases, and dozens of other survival and angiogenic proteins are switched on simultaneously. In the retina, a tissue with one of the highest metabolic oxygen demands in the body, this ancient oxygen-sensing system becomes a double-edged sword.

The review systematically dissects how this switch malfunctions in the three leading causes of vision loss. In diabetic retinopathy, chronic hyperglycemia does more than starve retinal tissue of oxygen through capillary dropout; it also stabilizes HIF-alpha directly through oxidative stress and inflammatory pathways, even in relatively well-oxygenated regions. In retinopathy of prematurity, the premature infant’s retina, still developing its vascular supply in a hyperoxic incubator environment, undergoes vaso-obliteration followed by a hypoxic phase in which HIF-driven VEGF floods the tissue and spawns disorganized, leaky vessels that can detach the retina. In neovascular age-related macular degeneration, the choroidal vasculature beneath the macula invades the retinal pigment epithelium in response to a hypoxic, inflamed, and drusen-laden microenvironment, with HIF orchestrating the choroidal neovascular membranes that hemorrhage and scar. In each disease, the authors emphasize, HIF integrates diverse upstream insults into a common downstream effector network.

Herein lies the problem with anti-VEGF therapy, the current standard of care. Drugs such as ranibizumab, aflibercept, and bevacizumab neutralize a single growth factor, and they transformed outcomes when introduced, saving the sight of millions. Yet a substantial fraction of patients respond inadequately or lose efficacy over time, a phenomenon the review frames as a structural limitation rather than a pharmacological accident. Because VEGF is only one branch of the HIF-dependent program, blocking it leaves the master switch intact and free to compensate through alternative angiogenic pathways, including placental growth factor, angiopoietins, hepatocyte growth factor, and inflammatory cytokines. Resistance emerges not because the drug fails to bind its target but because the upstream transcriptional engine keeps running, producing ever more redundant signals. Titrating the conductor, the authors argue, is more rational than silencing a single instrument in the orchestra.

The translational centerpiece of the review is its survey of HIF-targeted drug candidates, with particular attention to two lead compounds: 32-134D and PX-478. Both are small-molecule inhibitors designed to suppress HIF-alpha accumulation or activity, and both have progressed through preclinical evaluation with encouraging ocular data. By damping the master switch itself, these agents promise broader therapeutic coverage than any single-ligand blockade, potentially addressing leakage, neovascular proliferation, and fibrotic remodeling in one stroke. The authors also survey strategies at the molecular level, including approaches that modulate prolyl hydroxylase activity, disrupt HIF dimerization, or interfere with co-activator recruitment, as well as emerging delivery platforms suited to the eye’s immune-privileged and anatomically constrained environment. The review frames these efforts as a pipeline moving from bench chemistry toward clinical validation, though it is careful to note that no HIF inhibitor has yet replaced anti-VEGF injection in routine retinal practice.

Perhaps the most scientifically nuanced section of the review concerns HIF’s dual identity. The same transcription factor that fuels pathological vessel growth is indispensable for physiological vascular development. Embryonic retinal vessels form under HIF guidance; the ordered sprouting, tip-cell migration, and anastomosis that build a functional capillary network depend on precisely calibrated HIF signaling. Complete, indiscriminate suppression of the pathway risks impairing wound healing, neuroprotection, and normal vascular maintenance, particularly in premature infants whose retinas are still under construction. The authors stress that therapeutic success will hinge on precision in three dimensions: timing, so that treatment is delivered when pathological signaling dominates; cell type, so that pathogenic stabilization of HIF in endothelial cells, pericytes, or retinal pigment epithelial cells is targeted without disabling protective programs elsewhere; and microenvironment, so that the specific mix of hypoxic, metabolic, and inflammatory cues driving disease in each patient is taken into account.

This framing carries real clinical weight for the millions of people affected by these conditions. Diabetic retinopathy remains a leading cause of blindness in working-age adults as global diabetes prevalence climbs; neovascular age-related macular degeneration threatens an aging population in which patients face years of monthly or bimonthly intravitreal injections and a meaningful minority derive limited benefit; and retinopathy of prematurity grows more relevant as neonatal intensive care expands in low- and middle-income countries. A therapy that acts upstream could, in principle, reduce injection frequency, overcome tachyphylaxis, and address the fibrotic late stages that anti-VEGF drugs handle poorly. The review’s synthesis suggests that the field’s long-standing focus on VEGF, while enormously productive, may have been a necessary but incomplete first act.

The authors are equally candid about limitations. HIF biology is pleiotropic: the factor regulates metabolism, erythropoiesis, cell survival, and immune function throughout the body, raising concerns about systemic toxicity if inhibitors escape the eye. Pharmacokinetics in the vitreous, the optimal molecular target within the HIF pathway, and the risk of interfering with physiological repair processes all remain open questions. Clinical evidence for compounds like 32-134D and PX-478 in ocular disease is still maturing, and the review explicitly calls for further work to define which patients, which disease stages, and which combinations with existing anti-VEGF agents will maximize benefit. The dual role of HIF means that the therapeutic window, while real, must be mapped with care rather than assumed.

What emerges from the review is less a single breakthrough than a reframing of the problem. Ocular neovascularization, in this account, is not a VEGF excess disease but a transcriptional state disease, in which a master oxygen sensor is locked in the on position by hypoxia, hyperglycemia, oxidative stress, and inflammation. Anti-VEGF therapy treats the loudest symptom; HIF-targeted strategies aim at the control circuit itself. If the lead candidates now moving through translational pipelines can deliver broad efficacy with an acceptable safety profile and the precision of timing, cell type, and microenvironment that the authors demand, the standard of care for blinding retinal disease could shift from repeated downstream blockade to durable upstream control. For patients facing a lifetime of injections or the prospect of irreversible vision loss, that would represent one of the most consequential advances in ophthalmology in a generation.

Subject of Research: The role of hypoxia-inducible factor in ocular neovascularization and HIF-targeted strategies to overcome anti-VEGF resistance

Article Title: Hypoxia-inducible factor as a master switch in ocular neovascularization: overcoming anti-VEGF resistance and exploring therapeutic prospects

Article References: Liu, T., He, Y.-F., Liao, Y.-F., Wu, X.-J., Zhang, Y.-P., Li, J., Liu, J.-X., Wang, T., Wu, Z.-X., & You, Z.-P. (2026). Hypoxia-inducible factor as a master switch in ocular neovascularization: overcoming anti-VEGF resistance and exploring therapeutic prospects. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08976-4

Image Credits: AI Generated

DOI: 10.1186/s12967-026-08976-4

Keywords: hypoxia-inducible factor, ocular neovascularization, anti-VEGF resistance, diabetic retinopathy, retinopathy of prematurity, age-related macular degeneration, angiogenesis, HIF inhibitors, 32-134D, PX-478, vascular leakage, translational medicine

The study, published in BMC Medical Imaging, introduces MDG-Net, a Multi-level Decoder and Multi-Attention Feature Fusion Network developed by Jiajia Ni, Jinxin Xu, Cheng Tong, Guqiang Li, and Jingyu Sun, with affiliations spanning Anhui Polytechnic University, the 724 Research Institute of CSIC, Binzhou Medical University, and the Chery Automobile research center. What makes the work remarkable is not merely that it performs well, but what it removes. For nearly a decade, the dominant architecture for medical image segmentation has been the U-shaped network, a design in which an encoder compresses an image into abstract features and a decoder reconstructs a detailed map, with so-called skip connections shuttling information directly from the early layers to the late ones. MDG-Net throws the skip connections away entirely, and the field should pay attention to why.

To understand the significance of that decision, it helps to understand what skip connections were supposed to do. In a U-shaped network, the earliest layers of the encoder capture fine, low-level details such as edges, textures, and thin structures, while deeper layers capture high-level semantic information about what those details collectively represent. Skip connections exist to carry the fine details forward, so that the decoder can use them when painting the final segmentation. In retinal images, where the target structures are hair-thin vessels against a noisy background, this handoff is critical. But it comes at a cost. The skip connections do not discriminate. They carry not only the useful fine detail but also a flood of irrelevant background information, and as the authors of the new study describe it, this leads to feature dilution, a condition in which the signal the network actually needs is swamped by noise it does not.

Feature dilution is not a trivial inconvenience. In vessel segmentation, the difference between a correctly traced capillary and a missed one can hinge on a handful of pixels, and when background noise contaminates the features that guide the decoder, thin vessels are precisely the structures that suffer first. They occupy few pixels, offer weak contrast, and are easily confused with shadows, lesions, or imaging artifacts. A network whose decoding pathway is diluted by irrelevant features will systematically under-detect the smallest vessels, and those smallest vessels are often where early disease signs appear. The research team’s insight was that instead of trying to filter the noise out of the skip connections, it might be better to eliminate the pathway altogether and rebuild the decoder so that it never needed the shortcut in the first place.

The first pillar of MDG-Net is the Multilevel-Decoder Structure, or MDS module. Rather than relying on skip connections to inject low-level features into the decoding stage, the MDS module uses those low-level features directly within the decoder itself, exploiting them to capture information at multiple scales as the segmentation map is progressively reconstructed. In practical terms, the decoder no longer passively receives a noisy parcel from the encoder; instead, it actively mines the low-level representations at each stage of decoding, recovering fine structural detail while gathering multi-scale context. This means the network can simultaneously reason about a vessel’s place in the global vascular tree and about the local pixel-level evidence that a thin branch exists, without the two streams of information contaminating one another along the way.

The second pillar is the Multi-Attention Feature Fusion module, or MAF. Attention mechanisms, which have transformed fields from language modeling to protein structure prediction, allow a network to dynamically weight which parts of its internal representation matter most for a given decision. The MAF module applies this principle to feature fusion, expanding the network’s receptive field, the region of the original image that influences any single output prediction, and sharpening its semantic representation learning. A larger receptive field means the network can judge whether a faint linear structure is a vessel by consulting a wider swath of surrounding context, such as whether that structure connects plausibly to the broader vascular network. Better semantic representation means the network’s internal concept of what a vessel is becomes more robust to the variations in lighting, contrast, and pathology that make real-world retinal images so messy.

The proof, as always, lies in the benchmarks. The researchers evaluated MDG-Net on five widely used public retinal vessel datasets: DRIVE, STARE, CHASE_DB1, IOSTAR, and LES-AV. These datasets collectively span the major imaging modalities and patient populations used in the field, from color fundus photographs to vascular images with differing resolutions and pathologies, and they have served for years as the proving grounds on which competing segmentation methods are measured. Across all five, MDG-Net outperformed state-of-the-art methods, achieving higher accuracy and higher AUC scores, the area under the receiver operating characteristic curve that captures how well a model distinguishes vessel pixels from background across all possible decision thresholds. Consistency across five heterogeneous datasets is a stronger signal than a single record-setting score, because it suggests the architecture is genuinely robust rather than finely tuned to one dataset’s quirks.

The implications reach well beyond ophthalmology. The U-shaped encoder-decoder paradigm that MDG-Net departs from underpins segmentation systems used throughout medical imaging, from tumor delineation in MRI scans to organ boundary detection in CT volumes. If the skip connections that nearly all of these systems share are a source of feature dilution, then the demonstration that a carefully designed decoder can recover fine detail without them opens a new design space for the entire discipline. The authors describe MDG-Net as a robust alternative to traditional U-shaped architectures, and the phrase is carefully chosen: the goal is not to win a single leaderboard but to offer a different template for building segmentation networks, one in which the decoder earns its detail rather than inheriting it through a noisy shortcut.

There are also practical reasons for optimism about adoption. The study used only publicly available datasets and involved no new studies with human participants or animals, which simplifies the path to replication. The source code has been released on GitHub, allowing other research groups to test, extend, and stress-test the architecture on their own data. The work was supported by the Shandong Provincial Natural Science Foundation, the Anhui University Natural Science Foundation, and related institutional programs, with the funding bodies reporting no role in the study design, data analysis, or manuscript preparation. The article is open access, published under a Creative Commons license that permits sharing and reproduction with appropriate credit, meaning the full technical details are available to any clinician or researcher curious enough to look.

None of this means the problem of retinal vessel segmentation is solved. Real-world deployment demands validation on data from scanners and populations beyond the public benchmarks, and the gap between benchmark performance and clinical reliability is one that medical AI has stumbled over repeatedly. But the core contribution of MDG-Net is conceptual as much as empirical: it identifies a structural flaw in the field’s default architecture, explains the mechanism of that flaw, and demonstrates a working remedy. In a discipline where progress often comes from adding more layers, more parameters, and more connections, there is something quietly radical about a network that gets better by removing them. For the millions of patients whose retinas will be screened in the years ahead, that kind of radical simplicity may prove to be exactly what clear vision requires.

Subject of Research: Deep learning architecture for retinal vessel segmentation in medical imaging

Article Title: A multi-level decoder and multi-attention network for accurate retinal vessel segmentation

Article References: Ni, J., Xu, J., Tong, C., Li, G., & Sun, J. (2026). A multi-level decoder and multi-attention network for accurate retinal vessel segmentation. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02813-2

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02813-2

Keywords: retinal vessel segmentation, MDG-Net, deep learning, encoder-decoder architecture, skip connections, attention mechanism, medical image segmentation, BMC Medical Imaging, diabetic retinopathy, feature dilution, neural networks, computer vision

Retinal thickness is one of the most important structural biomarkers in ophthalmology. Swelling of the retina, known as edema, is a hallmark of diabetic macular edema, one of the leading causes of vision loss in working-age adults, and thickness changes also track the progression of glaucoma, age-related macular degeneration, and other conditions. Clinicians measure it using optical coherence tomography, or OCT, which bounces near-infrared light into the eye to build depth-resolved cross-sectional images of the retinal layers. From those scans, software computes a thickness map centered on the fovea, the pit of sharpest vision. The problem is that OCT machines are costly, require trained operators, and remain unavailable in many rural clinics, screening programs, and low-resource health systems. Color fundus photography, by contrast, is fast, cheap, and widely deployed — but it captures only a two-dimensional surface view, with no direct information about depth or thickness.

The research team, led by Maryam Yahyaie and Reza AghaeiZadeh Zoroofi of the University of Tehran’s School of Electrical and Computer Engineering, together with ophthalmologists Alireza Ramezani and Zhale Rajavi of Shahid Beheshti University of Medical Sciences, asked a deceptively simple question: if a machine learns the statistical relationship between the surface appearance of the retina and its underlying thickness, could it infer the missing dimension? Their answer is a generative artificial intelligence framework they call the Anatomically Conditional Latent Diffusion Model, or AC-LDM, trained on paired datasets in which each fundus photograph is matched to the thickness map derived from the same patient’s OCT scan.

The technical heart of the system is a diffusion model, the same family of generative architectures that has transformed image synthesis in recent years. Diffusion models learn to create data by reversing a gradual noising process: during training, the model observes images corrupted by increasing amounts of random noise and learns to denoise them step by step. At generation time, starting from pure noise, it iteratively refines a random field into a coherent image that matches the conditions it is given — in this case, the fundus photograph. Rather than operating on raw pixels, which is computationally expensive, the framework works in a compressed latent space. A variational autoencoder first learns to squeeze each retinal thickness map into a compact perceptual representation that preserves essential structure while discarding irrelevant detail, and the diffusion process then unfolds within that efficient latent space.

What distinguishes AC-LDM from a generic image-to-image translator is its anatomical guidance. The researchers observed that a standard cross-attention module — the mechanism that lets a diffusion model consult the fundus photograph while generating the thickness map — treats all spatial locations equally, which can blur or distort modality-specific retinal structure. Their solution is a spatially weighted cross-attention module, or SW-CA, which injects anatomically informed spatial weights into the attention computation. Regions of the fundus image that carry more reliable information about retinal topography are given greater influence during generation, while less informative areas are down-weighted. The design draws on attention-based conditioning strategies developed in modern vision architectures, but adapts them to the specific geometry and physiology of the retina, encouraging the model to respect the biological correspondence between what a fundus camera sees and what OCT measures.

Evaluating generative medical images is notoriously difficult, because a plausible-looking output is not necessarily an accurate one. The team therefore benchmarked AC-LDM against representative convolutional, generative adversarial, and transformer-based synthesis approaches using a battery of established metrics. Peak signal-to-noise ratio, or PSNR, quantifies pixel-level fidelity between the synthesized and true thickness maps; the structural similarity index, or SSIM, captures preservation of local structure such as the foveal depression and perifoveal rings. Two perceptual and distributional measures round out the picture: LPIPS, the learned perceptual image patch similarity, which correlates with human judgments of visual difference, and FID, the Fréchet Inception distance, which compares the statistical distribution of generated images against real ones.

The results were striking. AC-LDM achieved a PSNR of 30.88 decibels and an SSIM of 0.871, indicating both high pixel-level agreement and strong structural fidelity with the ground-truth thickness maps. On the perceptual side, it recorded an LPIPS of 0.105 and an FID of 29.5, both lower — and therefore better — than the baseline models it was compared against. Lower LPIPS means the synthesized maps look more like real ones to learned perceptual features, while the lower FID indicates that the overall population of generated maps is statistically closer to authentic retinal topography. Together, these numbers suggest that the anatomically guided diffusion approach does not merely produce attractive images but captures genuine structural information about the retina.

The implications extend well beyond a single laboratory benchmark. If thickness maps can be reliably synthesized from fundus photographs, screening programs for diabetic retinopathy — which already rely heavily on fundus photography in community and telemedicine settings — could gain a proxy measure of macular edema without deploying OCT hardware. The framework also opens a door for computational ophthalmology more broadly: large-scale fundus image archives, collected over decades and numbering in the millions, could potentially be retrofitted with synthesized thickness information, enabling retrospective studies of retinal structure at a scale OCT has never achieved. Cross-modal synthesis of this kind may likewise support data augmentation, privacy-preserving data sharing, and the training of downstream diagnostic models.

The authors are careful, however, to frame the work as a research advance rather than a clinical tool. Their own conclusion notes that clinical validation with thickness-specific error metrics — such as mean absolute error, root mean squared error, and agreement measures for central foveal thickness under the standard ETDRS grid — remains necessary before any translational use. Image-fidelity and perceptual scores, while encouraging, do not by themselves prove that synthesized thickness values are accurate enough to guide treatment decisions such as initiating anti-VEGF injections for macular edema. The study was approved by the Research Ethics Committee of Shahid Beheshti University of Medical Sciences, conducted under the Declaration of Helsinki, and used fully anonymized retrospective imaging data hashed to comply with HIPAA and GDPR requirements.

Still, the trajectory is clear and compelling. Diffusion models have already conquered natural image generation, and this study demonstrates how domain knowledge — in this case, the anatomy of the retina — can be woven into their attention mechanisms to solve a genuinely medical problem. The work, which received no specific funding and is published open access, points toward a future in which the humble fundus camera, already ubiquitous from Tehran to rural screening vans, becomes a window not just onto the surface of the retina but into its hidden third dimension. For the hundreds of millions of people at risk of retinal disease worldwide, many of whom live where OCT machines may never reach, that window could make the difference between blindness detected too late and sight preserved in time.

Subject of Research: Cross-modal synthesis of retinal thickness maps from color fundus photographs using an anatomically guided latent diffusion model

Article Title: Latent diffusion with anatomical guidance for retinal thickness map synthesis

Article References: Yahyaie, M., Zoroofi, R. A., Ramezani, A., & Rajavi, Z. (2026). Latent diffusion with anatomical guidance for retinal thickness map synthesis. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02792-4

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02792-4

Keywords: artificial intelligence, diffusion models, retinal thickness map, color fundus photography, optical coherence tomography, medical imaging, ophthalmology, deep learning, cross-modal synthesis, macular edema, diabetic retinopathy, computational ophthalmology

The surgery can be sight-saving, but outcomes vary dramatically from patient to patient. Some regain useful vision and keep it for years, while others deteriorate despite technically successful operations. A new study from Shanxi Eye Hospital in China, published in Health Science Reports, tackles this uncertainty head-on by building a machine learning model that predicts, before and during surgery, which patients are likely to end up with poor long-term vision. The work, led by Xiaolu Wang and colleagues, draws on 609 eyes from 609 patients who underwent their first vitrectomy for proliferative diabetic retinopathy between January 2022 and January 2025, each followed for at least twelve months after the operation.

The research team defined a good visual prognosis as an improvement of at least 0.3 units on the logarithm of the minimum angle of resolution scale, a standard measure of best-corrected visual acuity, by the final follow-up visit. Patients whose vision worsened by more than 0.3 units, or who failed to improve by that margin, were classified as having a poor prognosis. This twelve-month endpoint matters because previous studies have shown that visual acuity tends to stabilize after the first year following vitrectomy, making it a meaningful window for judging the true success of the intervention. The team even assigned numerical values to eyes that could only count fingers, perceive hand motion, or detect light, ensuring that even the most severely affected patients could be graded consistently.

Before any modeling began, the researchers confronted a challenge familiar to anyone working with real-world medical records: missing data. Of the 52 candidate variables collected from the hospital’s electronic medical record system, body mass index had the highest proportion of missing values at 19.2 percent. Rather than discarding incomplete records, the team used multiple imputation, a statistical technique that fills in gaps by drawing on the relationships among the observed variables. They then split the cohort randomly into a training set of 487 patients and a validation set of 122, verifying through descriptive statistics that the two groups were comparable across demographic, surgical, and biochemical characteristics.

Variable selection proceeded in two stages. First, the team applied least absolute shrinkage and selection operator regression, a technique that penalizes model complexity and drives the coefficients of uninformative variables toward zero. As the penalty parameter converged to 0.02910, seven candidate predictors survived the cut: age, renal insufficiency, preoperative iris neovascularization, the type of tamponade used during surgery, serum alkaline phosphatase, indirect bilirubin, and serum gamma-glutamyl transferase. These candidates then entered a binary logistic regression, where four emerged as statistically significant. Renal insufficiency carried an odds ratio of 6.932, meaning patients with impaired kidney function faced nearly seven times the odds of a poor visual outcome. Preoperative iris neovascularization was even more ominous, with an odds ratio of 7.674. Silicone oil tamponade doubled the risk at an odds ratio of 2.799, while higher indirect bilirubin levels appeared protective, with each unit increase associated with roughly a 10 percent reduction in the odds of poor prognosis.

With these predictors in hand, the researchers trained six different machine learning algorithms: decision tree, random forest, support vector machine, multilayer perceptron, logistic regression, and Light Gradient Boosting Machine, known as LightGBM. Each model was tuned through grid search and ten-fold cross-validation, a procedure that repeatedly partitions the training data to ensure the model’s performance is not a fluke of any single split. On the held-out validation set, the differences between algorithms became clear. The support vector machine, despite respectable training performance, saw its area under the receiver operating characteristic curve collapse to 0.616 on unseen data, a classic signature of overfitting. The multilayer perceptron fared somewhat better at 0.753 but still showed a troubling gap between training and test performance. Random forest reached 0.767 but suffered from weak recall and accuracy, while decision tree and logistic regression posted areas under the curve of 0.717 and 0.783 respectively.

LightGBM emerged as the clear winner, achieving an area under the curve of 0.786 on the validation set along with the best recall score during cross-validation at 0.979. Calibration curves confirmed that the model’s predicted probabilities tracked observed outcomes well in both the training and validation cohorts. Decision curve analysis, which quantifies the clinical net benefit of using a model at various risk thresholds, showed that LightGBM outperformed both a strategy of treating all patients and one of treating none, at least across low-to-moderate risk thresholds. The authors candidly note that the model’s net benefit fluctuated in the moderate threshold range, approaching zero at certain points, which they attribute to sparse sample distribution in that interval or reduced calibration, a reminder that even the best models have limits.

Perhaps the most clinically valuable contribution is the study’s effort to open the black box. Machine learning models are often criticized for being inscrutable, making it hard for physicians to understand why a particular prediction was made. To address this, the team applied SHAP, or SHapley Additive exPlanations, a framework borrowed from game theory that assigns each feature a contribution value for every individual prediction. The SHAP analysis confirmed the regression findings: renal insufficiency, silicone oil tamponade, and preoperative iris neovascularization all pushed predictions toward poor prognosis, while higher indirect bilirubin pushed toward good prognosis. The researchers also built a web-based calculator that lets clinicians enter a patient’s values and receive an instant estimate of poor-prognosis risk, translating the algorithm into a practical bedside tool.

The biological stories behind the risk factors are compelling. Renal insufficiency reflects systemic vascular vulnerability: reduced kidney function impairs the clearance of uremic compounds that amplify inflammation and oxidative stress in the retina, raising levels of vascular endothelial growth factor, the very molecule that drives abnormal blood vessel growth in diabetic eye disease. Iris neovascularization, present in about 65 percent of proliferative diabetic retinopathy patients according to prior research, signals severe retinal ischemia; roughly 20 percent of these patients progress to neovascular glaucoma, a painful and often blinding condition. The association between silicone oil tamponade and poor outcomes requires careful interpretation, since surgeons reserve oil for the most complex cases such as tractional retinal detachment, meaning the oil may be a marker of disease severity rather than a direct cause of visual decline, though the authors note it can also dissolve lipophilic macular pigments and exert chronic mechanical pressure on the retina.

The protective role of indirect bilirubin is the study’s most novel finding. Bilirubin, long dismissed as merely a waste product of hemoglobin breakdown, is now recognized as one of the body’s most potent endogenous antioxidants, capable of neutralizing free radicals and suppressing oxidative reactions. Slightly elevated levels may reduce intracellular oxidative stress, improve insulin sensitivity, and regulate glucose metabolism, all of which could slow diabetic complications. The authors are appropriately cautious, calling for future analyses of nonlinear dose-response relationships, mediation by anti-inflammatory biomarkers, and sensitivity analyses to rule out confounding by outlier values. They also acknowledge the study’s main limitations: the data came from a single hospital’s retrospective records, and the model was validated only internally. Multi-center prospective studies will be needed before the tool can be widely deployed. Still, the work points toward a future in which ophthalmologists can identify high-risk patients before they reach the operating table, intervene earlier on kidney function and neovascular disease, and tailor follow-up care to those who need it most, potentially preserving sight for thousands of people who would otherwise face preventable blindness.

Subject of Research: A machine learning prediction model for long-term visual outcomes after vitrectomy in proliferative diabetic retinopathy

Article Title: Construction and Validation of a Long‐Term Visual Prognosis Prediction Model for Proliferative Diabetic Retinopathy After Vitrectomy Based on Machine Learning

Article References: Wang, X., Shi, J., Gao, Y., Li, T., Han, X., Guo, L., Jia, T., & Wang, Y. (2026). Construction and Validation of a Long‐Term Visual Prognosis Prediction Model for Proliferative Diabetic Retinopathy After Vitrectomy Based on Machine Learning. Endocrinology, Diabetes & Metabolism, 9(5), Article e70314. https://doi.org/10.1002/edm2.70314

Image Credits: AI Generated

DOI: 10.1002/edm2.70314

Keywords: diabetic retinopathy, vitrectomy, machine learning, LightGBM, visual prognosis, SHAP, iris neovascularization, renal insufficiency, indirect bilirubin, silicone oil tamponade, predictive modeling, ophthalmology

The research, published in BMC Endocrine Disorders, examined 1,090 hospitalized patients with type 2 diabetes and asked a deceptively simple question: does anemia change how well a lesser-known blood sugar marker, albumin-corrected fructosamine, predicts diabetic retinopathy? The answer was a clear yes. Among patients with anemia, higher fructosamine levels were dramatically more strongly associated with retinopathy than among patients without it, and the marker outperformed HbA1c in identifying those at risk of eye damage.

To understand why this matters, it helps to look at the chemistry behind these tests. HbA1c measures glucose molecules that have permanently bonded to hemoglobin, the oxygen-carrying protein inside red blood cells. Because red blood cells live for roughly three to four months, the test reflects average glucose exposure over that window. But that same biology creates a vulnerability: anything that alters red blood cell survival distorts the reading. Anemia, whether from iron deficiency, chronic disease, or kidney dysfunction, shortens red cell lifespan, which can falsely lower HbA1c and make blood sugar control appear better than it truly is.

Fructosamine takes a different route. It measures glucose bonded to serum proteins, chiefly albumin, which circulate in the blood for only about two to three weeks. That shorter window makes fructosamine a gauge of recent glycemic control, and because it does not depend on red blood cells, it is theoretically immune to the distortions that plague HbA1c in anemic patients. The catch is that fructosamine levels rise or fall with albumin concentrations, so researchers typically correct the value against serum albumin, producing the albumin-corrected fructosamine, or AlbF, used in this study.

Diabetic retinopathy, the condition both markers are being asked to predict, remains one of the most feared complications of diabetes. It develops when chronically elevated glucose damages the delicate blood vessels of the retina, causing them to leak, swell, and eventually grow abnormally. It is a leading cause of blindness in working-age adults, and its earliest stages are silent, detectable only through eye examinations. Finding reliable blood-based markers that signal elevated retinopathy risk could help clinicians intensify monitoring and treatment before irreversible damage occurs.

In the new study, retinopathy was present in 459 of the 1,090 participants, or 42.1 percent, while 366 participants, or 33.6 percent, met World Health Organization criteria for anemia. Using multivariable logistic regression to account for confounding factors, the researchers found that among anemic patients, each 10 μmol/g increase in AlbF was associated with nearly double the odds of having retinopathy, with an adjusted odds ratio of 1.98 and a 95 percent confidence interval of 1.42 to 2.76, a highly statistically significant result. Among patients without anemia, the association was considerably weaker, with an odds ratio of 1.30.

The team went beyond simple comparison, formally testing whether anemia modified the relationship on both multiplicative and additive scales, the two frameworks statisticians use to detect interaction between variables. On the multiplicative scale, the interaction was significant, with a P value of 0.014. On the additive scale, the relative excess risk due to interaction was 2.88, the attributable proportion was 0.49, and the synergy index was 2.42, all indicating that the combined presence of anemia and elevated fructosamine confers risk beyond what either factor alone would predict. In practical terms, nearly half of the retinopathy risk in anemic patients with high AlbF could be attributed to this interaction.

Perhaps the most clinically consequential finding came from the head-to-head comparison of the two glycemic markers within the anemic subgroup. Adding AlbF to a base statistical model significantly improved the model’s ability to discriminate between patients with and without retinopathy, raising the area under the receiver operating characteristic curve from 84.43 percent to 87.07 percent, a difference that reached statistical significance at p equal to 0.012. Adding HbA1c, by contrast, provided no significant improvement. AlbF also improved risk classification beyond HbA1c, with a categorical net reclassification improvement of 0.109, a continuous NRI of 0.461, and an integrated discrimination improvement of 0.039, all statistically significant.

These numbers translate into a concrete clinical message. In anemic patients with type 2 diabetes, a group in which HbA1c is known to be unreliable, albumin-corrected fructosamine appears to carry real, incremental information about retinopathy risk that HbA1c cannot supply. The authors conclude that AlbF deserves consideration as a complementary glycemic marker in this population, not a wholesale replacement for HbA1c, but an additional lens through which to view a patient’s metabolic status when the standard test may be distorted.

The study’s cross-sectional design imposes important limits. Because blood was drawn and eye status assessed at the same time, the findings show association rather than proven prediction over time, and they cannot establish that elevated fructosamine causes retinopathy or that lowering it would protect the retina. The participants were all hospitalized patients at a single Chinese institution, which may limit generalizability to outpatient populations or other ethnic groups. Still, the large sample size, the rigorous dual-scale interaction analysis, and the formal reclassification statistics give the findings unusual statistical weight for a single-center study, and they align with a growing body of literature on the pitfalls of HbA1c in hematological disorders.

For clinicians, the practical takeaway is to think twice before interpreting an HbA1c result in a diabetic patient with known or suspected anemia, and to consider fructosamine-based testing as a cross-check, particularly when retinopathy screening decisions hang in the balance. For researchers, the results open a clear path forward: prospective studies tracking whether AlbF measured today predicts retinopathy progression years later, and trials examining whether correcting anemia restores concordance between the two markers. As diabetes rates continue to climb globally, refining the tools used to judge blood sugar control is not an academic luxury but a matter of preserved eyesight.

Subject of Research: The effect of anemia on the association between albumin-corrected fructosamine and diabetic retinopathy in type 2 diabetes

Article Title: Anemia modifies the association between albumin-corrected fructosamine and diabetic retinopathy in type 2 diabetes: a cross-sectional study

Article References: Anemia modifies the association between albumin-corrected fructosamine and diabetic retinopathy in type 2 diabetes: a cross-sectional study. (n.d.). https://doi.org/10.1186/s12902-026-02571-w

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02571-w

Keywords: diabetic retinopathy, albumin-corrected fructosamine, HbA1c, anemia, type 2 diabetes, glycemic biomarker, effect modification, cross-sectional study, retinopathy screening, serum albumin, hemoglobin, BMC Endocrine Disorders

The research, led by Yingying Zhu, Chun Jiang, Xiuhui He, Xiang Gao, and corresponding author Zhengxuan Jiang of the Department of Ophthalmology at The Second Affiliated Hospital of Anhui Medical University, describes a previously underappreciated signaling chain that connects the metabolic chaos of diabetes to the physical collapse of the retinal endothelial barrier. At the heart of the discovery is histone lactylation, a relatively recently identified epigenetic mark in which lactate-derived lactyl groups are chemically attached to lysine residues on histone tails. Rather than being an inert byproduct of metabolism, lactate in this context acts as a signaling molecule that reshapes which genes are switched on inside retinal blood vessel cells.

To dissect the mechanism, the team assembled evidence from multiple complementary systems. They examined human epiretinal membranes and fibrovascular membranes obtained from patients with proliferative diabetic retinopathy, retinal tissue from diabetic rats, and retinal endothelial cells grown under diabetic-like conditions. Across all of these models, a consistent pattern emerged: where lactate accumulated, protein lactylation rose, and one particular mark—lactylation at lysine 9 of histone H3, abbreviated H3K9la—stood out as prominently elevated under diabetic conditions. This convergence across human tissue, animal models, and cultured cells gave the finding a robustness that single-model studies often lack.

The critical question was what H3K9 lactylation actually does inside these endothelial cells. Histone modifications of this kind generally work by altering the physical state of chromatin, the complex of DNA and protein that packages the genome. When specific histone residues are acetylated or lactylated, the chromatin at nearby genes tends to loosen, granting the transcriptional machinery access and boosting gene expression. The researchers found that lactate-driven H3K9la became enriched at the promoter region of the PTK2 gene, which encodes focal adhesion kinase, a well-known regulator of cell adhesion, migration, and survival. With the promoter epigenetically opened up, PTK2 transcription increased, and levels of the phosphorylated, active form of the kinase climbed in parallel.

From there, the story moves from the nucleus to the cytoskeleton. Activated PTK2 was found to associate with FMNL2, a formin-family protein that governs the assembly of actin filaments, and this association was linked to increased tyrosine phosphorylation of FMNL2 itself. The consequence was a cascade of cytoskeletal remodeling inside the endothelial cells: the internal scaffolding of the cells reorganized in a way that destabilized VE-cadherin, the adhesive molecule that stitching neighboring endothelial cells together at adherens junctions. When VE-cadherin junctions falter, the endothelial sheet loses its seals, permeability rises, and fluid and proteins leak across the barrier. In the retina, that leakage translates directly into macular edema and progressive vision impairment.

What makes this axis scientifically compelling is that it forges a direct line from metabolism to cell structure through epigenetics. Diabetic tissue is known to be lactate-rich, a product of altered glucose metabolism and hypoxic stress. The study shows that this excess lactate does not merely fuel inflammation or oxidative damage indirectly; it physically marks the chromatin of barrier-regulating genes, amplifies a kinase–formin signaling module, and dismantles the junctions that hold the retinal vasculature together. In effect, a metabolic byproduct of diabetes becomes an epigenetic instruction that tells blood vessel cells to let go of each other.

Just as importantly, the research demonstrates that the damage is not irreversible in experimental settings. The team showed that pharmacologically reducing lactate production, inhibiting the catalytic activity of CBP/p300—the histone acetyltransferase enzymes responsible for writing lactylation marks—blocking PTK2 activity, or knocking down FMNL2 all attenuated endothelial barrier defects and reduced retinal vascular leakage. Each of these interventions targets a different rung on the same ladder, and the fact that several independent points of disruption produce protective effects strengthens the causal interpretation of the pathway and opens multiple potential angles for therapy.

The therapeutic implications are considerable. Existing treatments for diabetic retinopathy, such as anti-VEGF injections and laser photocoagulation, address downstream consequences of vascular dysfunction rather than the metabolic and epigenetic drivers of barrier failure. If the lactate–H3K9la–PTK2–FMNL2 axis can be safely modulated in patients—for example, by limiting lactate accumulation, tuning histone lactylation, or inhibiting focal adhesion kinase signaling locally in the eye—clinicians might one day intervene earlier in the disease process, before irreversible vascular damage takes hold. PTK2 inhibitors already exist in oncology research, and CBP/p300 catalytic inhibitors are under active investigation in multiple fields, meaning that repurposing strategies could accelerate translation.

The study also adds to a fast-growing body of literature on lactylation as a regulatory modification. Since histone lactylation was first described as a link between cellular metabolism and gene regulation, researchers have implicated it in macrophage polarization, tumor biology, fibrosis, and neural inflammation. The new work extends this framework to the vascular endothelium of the eye, suggesting that lactylation may be a general mechanism by which metabolically stressed tissues lose barrier integrity. Given that barrier failure is central to conditions ranging from sepsis to diabetic kidney disease, the conceptual reach of these findings may extend well beyond ophthalmology.

Caveats remain, as they always do at this stage of research. The pharmacological interventions were tested in experimental and preclinical systems, and the leap from rat retinas and cultured endothelial cells to human therapy will require careful validation, dosing studies, and safety assessment. Human tissue samples from proliferative diabetic retinopathy show the molecular signature, but they represent an advanced stage of disease; whether earlier interventions along this axis prevent progression is a question for future longitudinal work. Still, the identification of a defined metabolic–epigenetic–signaling pathway underlying inner blood–retinal barrier breakdown represents a genuine conceptual advance, one that reframes diabetic retinopathy not simply as a disease of damaged vessels, but as a disease of miswritten chromatin in the cells that guard the eye.

Subject of Research: Lactate-induced H3K9 histone lactylation disrupting the inner blood–retinal barrier via the PTK2–FMNL2 axis in diabetic retinopathy

Article Title: Lactate-induced H3K9 lactylation disrupts the inner blood–retinal barrier by activating the PTK2–FMNL2 axis in diabetic retinopathy

Article References: Zhu, Y., Jiang, C., He, X., Gao, X., & Jiang, Z. (2026). Lactate-induced H3K9 lactylation disrupts the inner blood–retinal barrier by activating the PTK2–FMNL2 axis in diabetic retinopathy. Cellular and Molecular Life Sciences. https://doi.org/10.1007/s00018-026-06448-y

Image Credits: AI Generated

DOI: 10.1007/s00018-026-06448-y

Keywords: diabetic retinopathy, inner blood–retinal barrier, histone lactylation, H3K9la, lactate, PTK2, FMNL2, VE-cadherin, endothelial permeability, focal adhesion signaling, epigenetics, retinal vascular leakage

The core insight behind UbiQVision is that explainable artificial intelligence, or XAI, and uncertainty quantification have usually been treated as separate problems, when in fact they are inseparable in high-stakes medicine. The dominant explanation technique for medical imaging is SHAP, short for SHapley Additive exPlanations, a game-theoretic method that assigns each pixel a contribution score indicating how much it pushed the model toward or away from a diagnosis. SHAP produces visually compelling heatmaps that clinicians find intuitive. But the method carries hidden assumptions. Standard SHAP formulations effectively treat features as independent, while pixels in medical images are strongly correlated. When the underlying data distribution is misspecified or estimated from small, biased samples, SHAP values can become unstable, producing misleading rankings of imaging biomarkers or spurious emphasis on artifacts. Clinicians, susceptible to automation bias, may over-trust visually appealing heatmaps that do not faithfully reflect the model’s true reasoning.

UbiQVision addresses this by unifying three mathematical disciplines into a single pipeline. First, the researchers constructed a heterogeneous ensemble of three distinct neural network architectures: a lightweight custom convolutional neural network, the widely used residual network ResNet-18, and a Vision Transformer pre-trained on ImageNet. Architectural diversity matters because it ensures the models’ errors are not perfectly correlated, a prerequisite for meaningful evidence fusion. Second, instead of averaging the ensemble’s predictions uniformly, the framework applies Bayesian meta-learning. Each model’s reliability is modeled as a random variable following a Dirichlet distribution, updated with validation performance scores such as F1 metrics. A temperature parameter controls how sharply the weighting favors the strongest model, and sampling from this posterior gives each model a probabilistic vote that rewards robust performers while preserving the influence of weaker models that may have learned strong local evidence.

The third and most novel component is the transformation of SHAP attributions into basic probability assignments within Dempster–Shafer evidence theory, a classical framework for reasoning under uncertainty. Using a hyperbolic tangent transformation scaled by a sensitivity parameter, the framework maps unbounded, real-valued SHAP scores into bounded evidential masses. Positive attributions become mass supporting the target diagnosis, negative attributions become mass supporting its negation, and any leftover mass is assigned to the universal set, representing total epistemic ignorance. Dempster’s rule of combination then fuses the weighted masses from all three models into pixel-level maps of belief, plausibility, and uncertainty. A conflict coefficient, computed during fusion, explicitly quantifies where the models disagree, rather than smoothing that disagreement away as conventional ensemble averaging does.

The resulting outputs map directly onto clinical concepts. The belief map marks regions where the ensemble has reached confirmed consensus, such as the dark, ring-like chromatin structures of a malaria parasite inside an infected red blood cell. The plausibility map captures the upper bound of what could be true, exposing internal conflict when, for example, the noisy ResNet model highlights random tissue as pathological while the other models disagree. The uncertainty map quantifies total ignorance: bright yellow regions signal that the model genuinely knows nothing, correctly covering empty slide background or out-of-distribution inputs, while dark purple regions indicate the model has sufficient evidence to decide. This explicit separation of confirmed disease, conflicting opinions, and insufficient data is precisely what standard softmax classifiers, which force every pixel into a category, cannot provide.

The team evaluated the framework across three publicly available medical imaging datasets spanning histology, neuroimaging, and ophthalmology. On the NIH malaria dataset of 27,558 balanced blood smear images, the Bayesian weighting identified the custom CNN as the primary expert with a posterior weight of roughly 0.37, and the fused belief maps performed what amounts to semantic segmentation of the parasite, filtering out the cell wall and cytoplasm as irrelevant background. Ten-fold stratified cross-validation showed highly consistent macro F1 scores: ResNet averaged 96.2 percent, with the custom CNN and Vision Transformer close behind at 95.7 percent. Local Lipschitz stability analysis confirmed that the SHAP attributions feeding the fusion were mathematically stable, with all three architectures scoring below 0.0012, indicating the maps reflect genuine features rather than unstable gradient noise.

The Alzheimer’s disease experiments revealed perhaps the most clinically resonant behavior. Using T1-weighted MRI scans graded across four dementia stages, the framework captured the non-linear progression of brain atrophy by modulating its evidential confidence with disease severity. In moderate dementia cases, positive attributions aligned precisely with enlarged ventricular boundaries, and the belief map showed dense, localized clusters of confirmed pathological evidence. For very mild dementia, where atrophy is subtle and easily confused with healthy aging, the uncertainty maps showed widespread high entropy, mirroring the genuine diagnostic difficulty that human radiologists face. Notably, the framework exposed a well-known weakness in the field: the very mild dementia class produced the highest mean fused uncertainty, correctly signaling that the ensemble was operating near the limits of its knowledge rather than masking that limitation behind a confident label.

On the diabetic retinopathy dataset from the EyePACS Kaggle competition, the framework faced its hardest test, a five-class ordinal grading problem with subtle transitions between severity levels. Here the custom CNN struggled, achieving a mean macro F1 of only 46.1 percent, while the Vision Transformer and ResNet reached 68.7 and 67.9 percent respectively. The framework adapted, and its uncertainty behavior tracked clinical reality: severe diabetic retinopathy, characterized by massive hemorrhages and extensive ischemia, elicited the lowest median uncertainty, while proliferative disease with its ambiguous, newly forming vascular anomalies produced the highest. Ablation studies across all three datasets confirmed that progressive Gaussian blur, which destroys anatomical structure, caused mean fused uncertainty to rise monotonically, demonstrating that the framework’s ignorance estimates genuinely track epistemic uncertainty arising from missing structural information.

Beyond the maps themselves, selective prediction risk-coverage analysis showed that UbiQVision provides superior uncertainty calibration compared with deep ensemble variance, Monte Carlo dropout, and integrated gradients baselines. On the malaria dataset, the framework maintained a residual error rate of zero up to roughly 35 percent coverage, while baseline methods exhibited dangerous overconfidence spikes at lower coverage levels. The framework is entirely post-hoc and model-agnostic at the ensemble level, requiring no modification to validated training pipelines, which distinguishes it from evidential deep learning approaches that demand specialized loss functions. The authors acknowledge real limitations: computational cost is substantial, with inference times of 0.55 to 1.03 seconds per image and peak memory demands of 7.5 to 7.7 gigabytes, and image resolution was constrained to 128 by 128 pixels for most models due to the memory requirements of pixel-wise SHAP computation. Shared blind spots among models trained on identical data could also undermine the uncertainty estimates under adversarial conditions.

Even so, the implications for safety-critical medical AI are considerable. By making the unknown unknowns visible, the framework allows clinical workflows to route high-confidence predictions for expedited validation while directing uncertain or contested cases to expert review, a distinction directly relevant to regulatory requirements under the EU AI Act, which mandates transparency, robustness, and explainability in high-risk medical systems. The researchers envision extending the evidential fusion to multi-modal and longitudinal settings, tracking belief and ignorance at the patient level over time, and using the uncertainty outputs to drive active learning. The code is publicly available on GitHub, and the framework’s deeper contribution may be conceptual: it reframes medical AI from a binary classifier that masquerades confidence as certainty into a risk assessment tool that communicates, pixel by pixel, exactly how much it knows, how much it doubts, and where it is simply guessing.

Subject of Research: Uncertainty-aware explainable AI framework for reliable deep learning ensembles in medical imaging

Article Title: UbiQVision: Spatial Dempster-Shafer fusion of XAI attributions for reliable deep vision ensembles

Article References: Dubey, A., Anžel, A., İlgen, B., & Hattab, G. (2026). UbiQVision: Spatial Dempster–Shafer fusion of XAI attributions for reliable deep vision ensembles. Machine Learning with Applications, 25, Article 101000. https://doi.org/10.1016/j.mlwa.2026.101000

Image Credits: AI Generated

DOI: 10.1016/j.mlwa.2026.101000

Keywords: explainable AI, uncertainty quantification, Dempster–Shafer theory, medical imaging, deep learning ensembles, SHAP, Bayesian meta-learning, malaria detection, Alzheimer's disease, diabetic retinopathy, Vision Transformers, clinical decision support

The study, led by Paizhalaiti Abudureheman and corresponding author Ming Liu, enrolled 1,940 individuals hospitalized with type 2 diabetes between 2018 and 2025. Of these, 1,116 had been diagnosed before their 40th birthday, a group the researchers classified as early-onset disease, while 824 had been diagnosed at age 40 or later, the late-onset group. Rather than simply comparing raw complication rates between the two groups, the team applied direct age standardization to the 2020 China Population Census, restricting the analysis to four age strata shared by participants aged 40 years or older. This methodological choice matters enormously. Early-onset patients are, by definition, younger, and younger people generally have fewer complications regardless of disease severity. Standardizing for age strips away that advantage and asks a sharper question: at the same age, who is worse off?

The answer, on several fronts, was unambiguously the early-onset group. After age standardization, the prevalence of atherosclerotic cardiovascular disease was 23.5 percent among early-onset patients compared with 19.0 percent in the late-onset group, a statistically significant difference. The gaps for microvascular complications were far larger. Age-standardized prevalence of diabetic nephropathy, the kidney-damaging complication that is a leading cause of end-stage renal disease, reached 32.5 percent in the early-onset group versus just 12.5 percent in the late-onset group. Diabetic retinopathy, which can progress to vision loss, showed a similarly dramatic pattern: 28.0 percent versus 10.4 percent. Both differences were highly significant. Only diabetic peripheral neuropathy, nerve damage typically presenting as numbness or pain in the extremities, failed to reach statistical significance after standardization, at 36.6 percent versus 33.2 percent.

What makes these numbers particularly sobering is the contrast with the crude, unadjusted data. Before standardization, the early-onset group actually appeared to have lower rates of cardiovascular disease and peripheral neuropathy, simply because they were younger. This is precisely the kind of distortion that can lull clinicians into underestimating risk in young patients. The age-standardized analysis reverses that picture entirely, revealing that when a 45-year-old with early-onset disease and a 45-year-old with late-onset disease stand side by side in a clinic, the one diagnosed young is more likely to have diseased coronary arteries, leaking kidney filters, and damaged retinal vessels.

The metabolic profile of the early-onset patients helps explain why. The study documented poorer glycemic control, greater insulin resistance measured by the homeostatic model assessment, more adiposity, more severe dyslipidemia, and higher serum uric acid in those diagnosed before 40. These are not independent curiosities; they form a coherent cluster of metabolic dysfunction that appears earlier and cuts deeper in young-onset disease. Early-onset type 2 diabetes is increasingly understood as a distinct phenotype, often driven by profound insulin resistance compounded by obesity, and in some individuals by an accelerated decline in beta-cell function that leaves them dependent on insulin therapy within years of diagnosis. The Northern China cohort reflects that biology in its laboratory values.

To probe the relationship between age at diagnosis and complications more finely, the researchers used restricted cubic splines, a statistical technique that models nonlinear associations without forcing the data into a straight line. These analyses revealed nonlinear relationships between age at diagnosis and both atherosclerotic cardiovascular disease and diabetic peripheral neuropathy, suggesting that the risk landscape shifts in complex ways across the diagnostic age spectrum rather than declining smoothly with each additional year of youth at onset.

Perhaps the most clinically actionable finding came from the multivariable analysis. The researchers entered nine prespecified covariates simultaneously into a logistic regression model examining factors associated with microvascular complications within the early-onset group. Five emerged as independent correlates: younger age at diagnosis, longer diabetes duration, higher body mass index, higher glycated hemoglobin, and higher serum uric acid. The uric acid signal was especially robust. Each 60-micromole-per-liter increase in serum uric acid, roughly the span between a typical normal value and a clearly elevated one, was associated with an adjusted odds ratio of 1.472 for microvascular complications, with a 95 percent confidence interval of 1.315 to 1.647. In practical terms, patients with elevated uric acid had nearly 50 percent higher odds of kidney, nerve, or eye complications within the early-onset group.

The authors are careful, appropriately, not to overclaim causality. Serum uric acid has been implicated in endothelial dysfunction, oxidative stress, and inflammation, and elevated levels are consistently associated with hypertension, chronic kidney disease, and metabolic syndrome. But the study is cross-sectional and retrospective, capturing a single moment in each patient’s disease course, and the researchers explicitly state that uric acid should be considered a complementary clinical correlate pending prospective validation. What the finding does argue for is attention: uric acid is inexpensive to measure, modifiable with existing drugs, and, if future interventional studies confirm the association, could become a genuine therapeutic target for protecting the microvasculature of young diabetic patients.

The broader implications extend well beyond Northern China. Type 2 diabetes diagnosed before age 40 is rising sharply across Asia, the Middle East, and increasingly in Western countries, driven by rising childhood and young-adult obesity. Epidemiological studies have repeatedly suggested that early-onset disease carries a disproportionate lifetime risk of complications and cardiovascular events, in part because a person diagnosed at 30 may live three or four decades with hyperglycemia and its metabolic consequences. This study’s age-standardized design strengthens that inference by showing that the excess burden is not merely an artifact of longer exposure measured in years but is visible when patients of the same chronological age are compared directly. The small vessels of a young body appear to be paying a disproportionately large price.

For clinicians, the message is a call to earlier and more aggressive assessment. The authors conclude that their findings support earlier metabolic and microvascular evaluation in young-onset type 2 diabetes, meaning screening for albuminuria, retinopathy, neuropathy, and cardiovascular risk should begin promptly after diagnosis rather than being deferred as it often is in younger patients perceived as low risk. For health systems confronting a generational wave of early-onset diabetes, the study is a warning that the complications once expected in the sixth and seventh decades of life are now emerging in people who have barely begun their careers. The microvascular clock, the data suggest, starts ticking faster and earlier than the calendar would imply, and the time to intervene is measured from the moment of diagnosis, not from the appearance of symptoms.

Subject of Research: Age-standardized comparison of microvascular and cardiovascular complications in early-onset versus late-onset type 2 diabetes

Article Title: Disproportionate microvascular burden emerging at younger ages in early- versus late-onset type 2 diabetes: an age-standardized analysis in Northern China

Article References: Abudureheman, P., Liu, T., Liu, Y., Liu, R., Wang, H., Shu, H., Shi, Q., Fan, Y., & Liu, M. (2026). Disproportionate microvascular burden emerging at younger ages in early- versus late-onset type 2 diabetes: an age-standardized analysis in Northern China. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02511-8

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02511-8

Keywords: early-onset type 2 diabetes, microvascular complications, diabetic nephropathy, diabetic retinopathy, serum uric acid, age-standardized prevalence, insulin resistance, cardiovascular disease, glycemic control, Northern China, BMC Endocrine Disorders, diabetes complications