Two Simple Numbers May Tell Apart Harmless and Dangerous Rare Stomach Tumors

A rare family of stomach tumors has long puzzled pathologists because some of them are essentially harmless while others can behave like genuine cancers, yet under the microscope they look strikingly similar. Now a team of Chinese researchers says the answer may lie in two deceptively simple measurements: the size of the lesion and the fraction of its cells that are actively dividing. In a letter to the editor published in Pathology & Oncology Research, a group based at Fujian Provincial Hospital and collaborating institutions reports a multicenter validation of a dual-parameter diagnostic rule that could change how these tumors are triaged in everyday clinical practice.

The tumors in question are collectively known as gastric epithelial neoplasms of fundic-gland mucosa lineage, abbreviated GEN-FGMLs. They arise from the specialized secretory cells of the fundic glands, the acid- and pepsin-producing factories of the stomach lining. Within this family sit two very different entities. Oxyntic gland adenoma, or OGA, is regarded as a low-atypia, indolent tumor that rarely, if ever, progresses to aggressive disease. Gastric adenocarcinoma of the fundic-gland type, or GA-FG, and the closely related fundic-gland mucosa type, GA-FGM, represent genuinely malignant differentiated tumors. Because all three share architectural and cytological features that overlap considerably, distinguishing them on a small biopsy has been one of the more stubborn problems in gastrointestinal pathology.

The new letter builds on an earlier study by the same team, published in 2024, that characterized the clinicopathological features of 37 GEN-FGMLs and proposed preliminary cutoff values. Those findings were subsequently presented at the 37th European Congress of Pathology in Vienna in September 2025. In the current communication, the authors extend and validate their proposal using an expanded multicenter cohort of 47 cases, comprising 24 oxyntic gland adenomas, 21 GA-FGs, and 2 GA-FGMs. The statistical signal in the expanded dataset was striking. Malignant-type lesions were significantly larger than OGAs, and their Ki67 proliferation indices were markedly higher, with medians of 7.5 percent for the carcinomas compared with 1.9 percent for the adenomas.

Ki67 is a well-established protein marker of cell proliferation. It is expressed by cells that have entered the active phases of the cell cycle and is visualized on tissue sections using immunohistochemistry, a staining technique that binds antibodies to the target protein and produces a brown signal visible under the microscope. The Ki67 index is calculated as the percentage of tumor cell nuclei that stain positive. In many tumor types, a high Ki67 index correlates with faster growth and worse prognosis. In the GEN-FGML family, the authors argue, this marker does something even more useful: it cleanly separates the indolent adenomas from the malignant carcinomas, which otherwise sit on a continuum of low cytological atypia.

To turn raw measurements into a usable diagnostic rule, the team applied receiver operating characteristic analysis, a standard statistical method for determining the threshold that best balances sensitivity and specificity. The analysis validated a Ki67 cutoff of 2.5 percent, achieving an area under the curve of 0.9476, with 90 percent sensitivity and 90.48 percent specificity. In plain terms, a lesion whose proliferating fraction exceeds 2.5 percent is very likely to be a carcinoma of fundic-gland lineage rather than an adenoma, and the rule misses relatively few true carcinomas while rarely mislabeling benign ones. The same analysis confirmed a size threshold of 4.5 millimeters, which achieved 94.74 percent sensitivity and 92.37 percent specificity, meaning lesions larger than this cutoff are also suspicious for the malignant entities.

Importantly, the authors did not stop at conventional accuracy metrics. They also performed decision curve analysis, a method that evaluates the net clinical benefit of a test across a range of threshold probabilities, weighting the harms of false positives against the harms of false negatives. This analysis showed that the Ki67 criterion delivered superior net benefit in moderate-risk ranges, reaching a normalized benefit of up to 0.9 at threshold probabilities between 0.3 and 0.6, while the size-based model performed best at lower risk thresholds, achieving 0.8 in the 0.2 to 0.4 range. The complementary strengths of the two parameters form the logical backbone of the proposed workflow: size is exquisitely sensitive, while Ki67 is highly specific.

The diagnostic pathway that emerges from these findings is refreshingly straightforward. The first step is measuring the macroscopic size of the lesion, which can be obtained from endoscopic images or from resection specimens without any additional testing. If a lesion measures 4.5 millimeters or less and its Ki67 index is 2.5 percent or below, it can be classified as an oxyntic gland adenoma with a specificity exceeding 90 percent. In that scenario, the authors suggest, conservative management or continued observation is appropriate, and immediate endoscopic resection may be unnecessary. Conversely, if either parameter exceeds its threshold, the lesion should be considered for endoscopic resection because it is likely to represent GA-FG or GA-FGM. When the two parameters disagree, for example a small lesion with a high Ki67 index or a large lesion with a low index, the authors recommend proceeding with resection, since combining the high sensitivity of size with the high specificity of Ki67 minimizes both missed diagnoses and unnecessary interventions.

One of the practical obstacles to using Ki67 in this context is well known to any practicing pathologist: inter-observer variability. Counting positively stained nuclei by eye is tedious and subjective, and the problem is magnified in small biopsy specimens where the available tumor area is limited and a handful of miscounted fields can shift the index substantially. To address this, the team employed an artificial intelligence-based nucleus recognition algorithm known as Domain-Specific Pruning, originally described at the MICCAI 2021 conference, which was specifically trained on Ki67-immunostained slides from GEN-FGMLs. According to the letter, the algorithm demonstrated high concordance with manual scoring by pathologists while enhancing reproducibility, effectively offering a standardized digital second opinion on the proliferative fraction.

The advantages of the dual-parameter workflow, the authors argue, extend beyond raw accuracy. First, it provides objective, quantitative criteria that are particularly valuable in biopsy samples, where the pathologist cannot assess whether tumor cells have invaded the submucosa, the layer beneath the stomach lining whose involvement is a key determinant of treatment. Second, it reduces inter-pathologist variability, a recognized source of diagnostic inconsistency for these rare tumors that few pathologists encounter frequently. Third, it is readily implementable in routine practice, because both size measurement and Ki67 immunohistochemistry are universally available in pathology laboratories, requiring no specialized equipment beyond what most hospitals already possess. This accessibility matters, because exotic molecular tests or advanced imaging modalities would be far harder to deploy in the community settings where most of these biopsies are actually read.

The authors are candid about the limits of their work. The cohort of 47 cases, while the largest yet assembled for this purpose, remains modest by the standards of common cancers, and the findings require prospective validation in larger, independent cohorts, particularly across different ethnic populations, since the prevalence and behavior of fundic-gland neoplasms may vary geographically. Nevertheless, the letter concludes on an optimistic note: the multicenter validation corroborates the diagnostic value of the Ki67 and size criteria originally proposed, and integrating Ki67 quantification into a simple dual-parameter pathway holds promise for standardizing the diagnosis of GEN-FGMLs, especially in biopsy settings. If larger studies confirm these results, the practical consequence could be significant: fewer patients with indolent oxyntic gland adenomas undergoing unnecessary endoscopic resections, and fewer early carcinomas of the fundic-gland type slipping through the diagnostic net because they looked deceptively benign on a small sample. In a field where a single millimeter and a single percentage point can redirect a patient’s entire management plan, that would be no small achievement.

Subject of Research: Diagnostic criteria using Ki67 index and lesion size to differentiate gastric fundic-gland lineage neoplasms

Article Title: A Letter to the Editor regarding the article “gastric epithelial neoplasm of fundic-gland mucosa lineage: representative of the low atypia differentiated gastric tumour and Ki67 may help in their identification”

Article References: Li, H., Zheng, L., Zhong, G., Chen, L., & Chen, X. (2026). A Letter to the Editor regarding the article “gastric epithelial neoplasm of fundic-gland mucosa lineage: representative of the low atypia differentiated gastric tumour and Ki67 may help in their identification”. Pathology & Oncology Research, 32, Article 1612588. https://doi.org/10.3389/pore.2026.1612588

Image Credits: AI Generated

DOI: 10.3389/pore.2026.1612588

Keywords: gastric tumors, Ki67 index, oxyntic gland adenoma, fundic-gland mucosa, pathology, biopsy diagnosis, immunohistochemistry, artificial intelligence, ROC analysis, decision curve analysis, endoscopic resection, gastric adenocarcinoma