Histopathology Metastasis Detection & pN Staging Simulator

Not medical advice. This page and its simulator are an educational illustration of a staging framework and a detection concept. They use a simplified simulated patch grid, not real tissue or real imaging data, and cannot diagnose or stage any real condition. Anyone with questions about an actual pathology result should speak with their treating physician or pathologist.

Why Lymph Nodes, and Why This Is Hard

Lymph nodes act as filtration checkpoints along the lymphatic system, and for many solid tumors, including breast cancer, they are usually the first place cancer cells travel to if they leave the original tumor site. Whether cancer is found in nearby lymph nodes, and how much, is one of the strongest available predictors of how a cancer is likely to behave and which treatments are appropriate, which is why the sentinel lymph node, the first node a tumor's lymphatic drainage reaches, is so often surgically removed and examined after a breast-cancer diagnosis. The examination itself is where the difficulty lies. A resected lymph node is sliced into thin sections, mounted on glass slides, and stained, typically with hematoxylin and eosin (H&E), so that cell nuclei and tissue structures become visible under a microscope. A pathologist then has to systematically scan the entire section at high magnification looking for tumor cells among the node's normal lymphoid tissue. A genuine metastatic deposit can be a large, obvious solid mass, but it can just as easily be a handful of scattered cells occupying a tiny fraction of the slide, easy to miss under time pressure across dozens of slides in a single case.

Whole-Slide Imaging and the CAMELYON Challenge

Digital pathology turns glass slides into whole-slide images: extremely high-resolution digital scans, often gigapixels in size, that can be viewed, annotated, and, crucially, processed computationally. That last part opened the door to a natural question: could a computer-vision algorithm, trained on enough labeled examples, learn to spot metastatic tumor regions in a whole-slide image on its own? CAMELYON (CAncer MEtastases in LYmph nOdes) was the public research challenge built to answer that question rigorously, using annotated whole-slide images of breast-cancer sentinel lymph-node sections. The findings, published by Bejnordi and colleagues, are well documented and worth stating precisely rather than overstating: the best-performing algorithms in the challenge achieved detection accuracy that matched or approached that of expert pathologists working under a fixed time constraint on the same slide set, and combining an algorithm's predictions with an independent pathologist review outperformed either the algorithm or the pathologist working alone. This did not establish that AI should replace pathologist diagnosis, and the researchers did not claim that. It established that AI-assisted screening, algorithms flagging regions for closer human review, is a credible and genuinely useful complement to expert judgment, not a replacement for it.

How a Patch-Based Detector Actually Works

A whole-slide image is far too large to feed into a model in one piece, so the standard approach, and the one this simulation illustrates directly, is to divide the image into a grid of small tiles, often called patches, and classify each patch independently as tumor or normal tissue. Stitching those per-patch predictions back together produces a heatmap over the whole slide, highlighting the regions most likely to contain tumor cells for a pathologist to examine more closely. This simulator's grid of 100 small tiles is a direct, simplified stand-in for that process: each tile represents a patch of simulated tissue, tiles are flagged tumor-positive or not, and the pattern of positive tiles is analyzed exactly the way a real pipeline would analyze a stitched heatmap, by looking for contiguous regions and measuring how large they are. Real systems are of course far more sophisticated, typically convolutional neural networks trained on thousands of expert-annotated patches, but the underlying logic, per-patch classification followed by spatial aggregation, is the same idea this simulation lets you manipulate directly with a slider.

The Real AJCC pN Staging Bands

Once a metastatic deposit is found, its size determines which of three real, clinically defined bands it falls into under the AJCC (American Joint Committee on Cancer) staging system, the same reference framework pathology reports actually use. Isolated tumor cells (ITC) describes single tumor cells or small clusters no larger than 0.2 mm. Nodes containing only ITCs are, by convention, excluded from the positive-node count used for N classification, though they are still recorded, because deposits this small have a different prognostic weight than a true metastasis. Micrometastasis covers deposits larger than 0.2 mm but not larger than 2.0 mm. This band is denoted pN1mi in a pathology report and does count as node-positive disease, distinct from ITCs. Macrometastasis covers any deposit larger than 2.0 mm, the band most people mean when they informally say a lymph node "has cancer in it," and it feeds into the pN1a and higher categories depending on how many nodes are involved. These size thresholds are not arbitrary: they reflect an accumulated body of clinical outcomes research linking deposit size to prognosis and to the likelihood that a patient benefits from additional treatment, which is exactly why a pathology report states a precise size category rather than simply "positive" or "negative."

This Simulator's Illustrative Scale Mapping

To make the real millimeter thresholds tangible on a small patch grid, this simulator states its scale mapping explicitly rather than leaving it implied. The 10×10 grid of 100 patches stands in for a simulated 2.0 mm × 2.0 mm field of a lymph-node whole-slide image, meaning each patch represents roughly 0.2 mm of tissue. That specific scale was chosen deliberately so the grid can illustrate all three real bands. A single isolated positive patch, surrounded by negative patches, lands right at the real 0.2 mm isolated-tumor-cell boundary. A contiguous cluster of a few adjoining patches lands in the 0.2-2.0 mm micrometastasis range. A deposit that grows to span the grid's full width or height approaches the real 2.0 mm boundary into macrometastasis. The simulator measures the largest connected group of tumor-positive patches, checks its width and height in simulated millimeters, and reports the illustrative pN category live, exactly mirroring how a real deposit's greatest linear dimension determines its real staging category. It is worth being explicit about the limits of this mapping: a 2 mm x 2 mm field of view is a small fraction of an actual lymph-node cross-section, which can be well over a centimeter across, and a 100-patch grid is a drastic simplification of a gigapixel whole-slide image processed by a trained neural network. The simulator is built to teach the size-threshold logic and the detection concept clearly, not to reproduce the full scale or complexity of a real digital-pathology pipeline.

Scattered vs. Clustered: Why Spatial Pattern Matters

The simulation's two spatial-pattern modes exist because real metastatic deposits are not just defined by how much tissue is involved, but by how that involvement is distributed. Isolated tumor cells are, true to their name, typically scattered as individual cells or tiny non-confluent clusters through otherwise normal-looking lymphoid tissue, exactly what the "scattered" mode places on the grid: positive patches kept apart from one another wherever possible. A genuine metastasis, by contrast, is a contiguous mass of tumor cells that grew from a common site, exactly what the "clustered" mode simulates by growing a single connected region outward from a random starting patch. This distinction is not just cosmetic in the simulator, it is diagnostically real: the same total number of tumor-positive patches can land in a completely different pN category depending on whether they are scattered or clustered, because the AJCC size thresholds are defined by the extent of a single contiguous deposit, not by the total tumor-positive area added up across separate, unconnected foci. Watching the pN badge change as you toggle the pattern at a fixed tumor-positive fraction is a direct, hands-on way to see why pattern, not just quantity, drives the real staging outcome.

Why AI-Assisted Screening Is a Real Answer to a Real Bottleneck

Return to the scale problem described earlier: a whole-slide image can represent thousands of individual high-power microscope fields, and a clinically meaningful deposit, especially a micrometastasis or isolated tumor cells, might occupy only a handful of those fields. Manually panning through every field at diagnostic magnification for every node from every patient, with sustained attention, is genuinely demanding work, and pathologist time is a limited, valuable resource in every health system. This is the well-documented motivation behind the CAMELYON line of research and the broader field of computational pathology: not to remove the pathologist from the loop, but to direct their attention efficiently, flagging the regions of a slide most likely to contain tumor so that expert review time is spent where it matters most. The finding that combined algorithm-plus-pathologist review outperformed either alone is the clearest evidence that this complementary framing, rather than an automation-replaces-expert framing, is the one actually supported by the published results.

Frequently asked questions

Try it live

Everything above runs in your browser — open Histopathology Metastasis Detection & pN Staging Simulator and change the tumor-positive fraction and spatial pattern while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.