The study, published in Medical & Biological Engineering & Computing, describes a deep learning pipeline built around a hybrid Transformer-convolutional neural network architecture called UDTransNet. The system ingests digitized whole slide images of gastric mucosa biopsies, identifies metaplastic glands at the pixel level, and then translates those segmentation results into a clinical severity grade. Crucially, the grading step follows the Sydney system, the internationally recognized standard that pathologists use to classify gastritis and intestinal metaplasia by estimating the proportion of tissue occupied by metaplastic glands.

The technical challenge the researchers faced is formidable. A single whole slide image can contain tens of thousands of glands scanned at resolutions exceeding a hundred thousand pixels per side, far beyond what any neural network can process in one pass. The team therefore employed multi-scale patch cropping, slicing each slide into overlapping tiles at several magnifications so the network can see both fine cellular texture and broader tissue architecture. UDTransNet then fuses the strengths of two complementary families of deep learning models: Transformers, whose self-attention mechanisms capture long-range spatial relationships across a tissue section, and convolutional neural networks, which excel at extracting precise local edge and texture features. The architecture’s learnable skip connections shuttle these multi-level features back into the decoder, allowing sharp, anatomically faithful segmentation boundaries.

The performance figures reported on the internal test set are striking. The segmentation model achieved a Dice coefficient of 0.9698, meaning that its automatically outlined metaplastic regions overlapped with expert annotations almost perfectly. When those segmentations were converted into Sydney-system severity grades, the model reached an accuracy of 0.8879 with a Kappa statistic of 0.85, a level of agreement conventionally interpreted as near-perfect concordance with expert pathologists.

Perhaps the most provocative comparison in the study is the one against human observers at different career stages. The model’s Kappa of 0.85 exceeded the diagnostic consistency measured between junior and intermediate pathologists, which ranged from 0.67 to 0.82. In practical terms, the algorithm agreed with expert diagnoses more reliably than less experienced human specialists agreed with one another. For clinical workflows in which biopsy volume is high and specialist time is scarce, that gap matters: it suggests a machine assistant could meaningfully reduce the inter-observer variability that has long plagued gastric precancerous lesion grading.

The framework also proved valuable at the other end of the diagnostic spectrum, in ruling disease out. When identifying completely normal tissue, the model achieved a recall of 92.1 percent and an F1 score of 95.9 percent. This ‘negative exclusion’ capability is clinically significant because patients whose biopsies are genuinely free of metaplasia can be triaged toward routine surveillance rather than intensive follow-up, freeing endoscopy and pathology resources for those at genuine risk. An automated system that reliably clears normal slides lets human experts concentrate their attention on the borderline and positive cases where judgment matters most.

What distinguishes this work from many prior AI pathology studies is its interpretability and its direct mapping to a clinical standard. Rather than outputting an opaque risk score, the system produces digital maps of metaplastic gland distribution across the slide and quantifies their area proportion in accordance with Sydney system thresholds. The authors describe this as the first establishment of a digital mapping between pathological morphological features and the Sydney grading criteria. Explanatory visualization techniques such as gradient-based localization, referenced in the paper, allow the model’s attention to be checked against recognizable histological landmarks, addressing a common barrier to clinical adoption of deep learning in medicine.

The clinical context underscores the urgency. Gastric cancer incidence and mortality remain high in East Asia, and the well-characterized Correa cascade of gastric carcinogenesis, from chronic inflammation through atrophy and intestinal metaplasia to invasive cancer, offers a window in which detection and intervention can change outcomes. Global meta-analyses cited by the authors indicate that gastric intestinal metaplasia is widespread, yet its histological grading still depends on subjective visual assessment, a process the study characterizes as inefficient and poorly reproducible. Prior computational efforts have largely focused on endoscopic images; this work moves the diagnostic burden onto the histopathology slide itself, where grading criteria are formally defined.

The implications extend beyond the pathology laboratory. An automated, standardized grading system could harmonize diagnoses across hospitals and countries, enabling more consistent risk stratification for endoscopic surveillance programs and cleaner data for epidemiological research and clinical trials. The authors note that their framework provides an efficient and interpretable solution for intelligent screening of gastric precancerous lesions, and the reported ethics approval from Fujian Provincial Hospital indicates the work is grounded in real clinical specimens. The research was supported by the National Natural Science Foundation of China and provincial science funds of Fujian Province, with contributions from institutions including South China Normal University, Tsinghua University Shenzhen International Graduate School, and Fuzhou University Affiliated Provincial Hospital.

Challenges remain before such systems become routine. External validation on slides from different scanners, staining protocols and patient populations will be essential, and regulatory pathways for AI-assisted diagnosis are still maturing. Yet the core result stands: a Transformer-CNN fusion network, trained on whole slide images and constrained by an internationally accepted grading standard, can match expert pathologists in segmenting and grading one of gastric cancer’s most important precursors. As digital pathology continues to replace glass slides with gigapixel images, tools like this one point toward a future in which every biopsy receives an instant, consistent, and explainable second opinion, one that never tires and never varies from one reading to the next.

Subject of Research: Automated deep learning segmentation and grading of gastric intestinal metaplasia in whole slide pathology images

Article Title: Gastric mucosa intestinal metaplasia segmentation and grading via transformer-CNN fusion architecture: an interpretable digital pathology diagnostic framework

Article References: Jin, Y., Zhu, L., Yan, X., Lin, L., Zhu, M., Yang, L., Han, S., Li, T., Zeng, Y., Ji, Y., & Li, H. (2026). Gastric mucosa intestinal metaplasia segmentation and grading via transformer-CNN fusion architecture: an interpretable digital pathology diagnostic framework. Medical & Biological Engineering & Computing. https://doi.org/10.1007/s11517-026-03658-1

Image Credits: AI Generated

DOI: 10.1007/s11517-026-03658-1

Keywords: gastric cancer, intestinal metaplasia, deep learning, whole slide imaging, digital pathology, semantic segmentation, Sydney system, Transformer, convolutional neural network, precancerous lesions, computer-aided diagnosis, UDTransNet

Axons, the long cable-like projections that carry signals between neurons, are fundamental to the structural integrity and function of the nervous system. Measuring their density and organization is central to understanding diseases such as multiple sclerosis, in which axonal loss drives progressive disability, as well as Alzheimer’s disease and traumatic brain injury. The Bielschowsky silver stain, first introduced in 1902, remains a workhorse of clinical neuropathology because it renders thin axons and neurofibrils dark against a paler background. But converting these images into reliable numbers has long been a bottleneck. Manual counting and stereological sampling are slow and scale poorly, while shortcuts such as optical density measurements are confounded by stain quality and by silver deposits in non-axonal tissue. Line-intersection counting, another common surrogate, depends on arbitrary line placement and local fiber orientation.

Classical image-processing approaches fare no better. Color-deconvolution techniques, which work reasonably well on brightly colored immunohistochemical stains, perform poorly on the monochromatic, heterogeneous contrast of silver impregnations. Existing deep learning pipelines for axon segmentation, such as AxonDeepSeg and DeepACSON, were built for electron microscopy, immunofluorescence, or myelin-specific stains, and none had been optimized for brightfield silver-stained histology. BasNet was designed specifically to fill that gap, trained entirely on manually annotated Bielschowsky-stained tissue.

The architecture is an attention U-Net, an encoder-decoder convolutional network in which so-called attention gates selectively emphasize task-relevant features while suppressing background noise. The encoder compresses the image through four hierarchical levels, expanding from 32 to 256 channels with each downsampling step, while the decoder reconstructs a full-resolution segmentation map. At every skip connection, an attention gate computes spatial attention coefficients that weight encoder features according to their relevance, a mechanism that proved particularly valuable in silver-stained tissue, where non-specific background silver deposition can trigger false positives in a standard U-Net. The complete network contains approximately 7.85 million trainable parameters and processes 256 by 256 pixel RGB patches, outputting a binary mask in which axon pixels are distinguished from background.

Training presented two classic challenges: extreme class imbalance, with axons occupying only about 30 percent of the pixels, and the need to recover very thin, elongated structures. The team addressed both with Focal Tversky Loss, a function that penalizes false negatives more heavily than false positives and applies a quadratic focusing factor that concentrates learning on difficult boundary regions. This asymmetry was deliberate. For axon density quantification, a missed axon directly reduces the density estimate, whereas slight over-segmentation is less harmful. The design shows in the results: mean recall of 0.754 consistently exceeded mean precision of 0.717 on the test set. An ablation study confirmed that the attention gates improved mean test Dice from 0.699 to 0.715, a small but consistent gain achieved at a cost of only about one percent additional parameters.

The training data were laborious to produce. Two medical doctors annotated 33 image tiles covering more than 25 million pixels, drawn from 26 whole-slide images spanning gray matter, white matter, demyelinated lesions, and background regions from the brains of four multiple sclerosis patients obtained through the German MS Brain Bank. Slides were prepared by different technicians at different times, deliberately capturing the real-world staining variability that plagues clinical material. The tiles were subdivided into more than 86,000 augmented training patches, and a leave-one-WSI-out cross-validation across twelve folds confirmed that no single training slide was disproportionately critical to performance, with Dice scores ranging narrowly between 0.687 and 0.725.

The most telling evaluation was a direct comparison with human variability. When two independent raters segmented the same two tiles, their mutual agreement, measured by the F1 or Dice coefficient, was 0.637 and 0.671. The model’s agreement with each individual rater ranged from 0.681 to 0.736, meeting or exceeding the human-to-human ceiling on both tiles. Comparable benchmarks in the literature, such as inter-rater Dice scores of around 0.78 for optic nerve axons and an average Jaccard index of 0.653 for nuclei segmentation, place BasNet squarely within the range of inter-observer disagreement. The authors conclude that the principal remaining source of error is annotation ambiguity rather than model limitation. On the held-out test set the model achieved a mean Dice of 0.717 and an intersection over union of 0.584, though performance varied widely across tiles, from perfect scores on a background tile to a low of 0.392 on a poorly stained slide, mirroring the difficulties that poor staining quality also poses to human annotators.

Beyond tile-level scores, the team demonstrated a pipeline that scales to whole slides. Aggregated segmentation masks produce spatially resolved maps of axon density at 70 micrometer resolution, revealing the expected anatomy: uniformly bright white matter tracts, a dimmer and more variable cortical ribbon, and visibly reduced density in multiple sclerosis lesions. A structural tensor analysis, built on image gradients and eigen decomposition, additionally yields maps of dominant fiber orientation and local coherency, a fractional anisotropy-like measure of directional alignment. Highly organized fiber bundles appear bright and smoothly colored, while crossing-fiber zones, cortex, and lesion borders show low coherency, consistent with known disruption of the tissue’s myeloarchitecture. These outputs provide quantitative readouts suitable for downstream statistical analysis and cohort comparison, in place of arbitrary sampling decisions.

As a benchmark comparison, the team trained nnU-Net v2, a widely used self-configuring segmentation framework, on the identical data. Their compact attention U-Net achieved a higher mean Dice on both the full eight-tile test set, 0.717 versus 0.555, and the seven axon-bearing tiles, 0.677 versus 0.634, while using roughly six times fewer parameters than nnU-Net’s self-configured network. The advantage was clearest on a white matter tile where axons run perpendicular to the cutting plane and appear as small cross-sectional profiles. The authors caution that with only eight test tiles, this comparison should be read as an honest benchmark rather than a definitive ranking.

The study has clear limitations. All tissue came from a single disease context, multiple sclerosis, and followed a single staining protocol, and all training data were digitized at one resolution of 0.273 micrometers per pixel. Preliminary tests at coarser resolutions looked visually convincing but were not quantitatively evaluated, and generalization to other silver impregnation methods or other laboratories remains unverified. The annotation set of 33 tiles is small, and the inter-rater ceiling was estimated from just two tiles. Nevertheless, the team has released the full pipeline as open-source software, with the trained model hosted on Hugging Face and the source code archived on Zenodo, making it immediately deployable on tiled images and whole-slide scans. The researchers envision future integration with multimodal registration frameworks that would align these histological orientation maps with ex vivo and in vivo MRI data, establishing a quantitative histological ground truth for microstructure imaging and opening a new bridge between the microscopic world of silver-stained axons and the macroscopic images used in clinical neurology.

Subject of Research: Automated deep learning segmentation of axons in Bielschowsky silver-stained histological brain sections

Article Title: BasNet: Attention U-Net-Based Automated Axon Segmentation in Bielschowsky Silver-Stained Histology

Article References: Schönenberger, L., Egli, L., Gkotsoulias, D., Stadelmann, C., & Granziera, C. (2026). BasNet: Attention U-Net-Based Automated Axon Segmentation in Bielschowsky Silver-Stained Histology. Neuroinformatics, 24(3), Article 59. https://doi.org/10.1007/s12021-026-09815-z

Image Credits: AI Generated

DOI: 10.1007/s12021-026-09815-z

Keywords: Bielschowsky silver stain, axon segmentation, attention U-Net, deep learning, computational pathology, multiple sclerosis, neuropathology, image segmentation, axon density, whole-slide imaging, Focal Tversky Loss, Neuroinformatics