Key concepts
HistoForge organizes an analysis as a module. You install a module, assign it to your images, and run it inside QuPath.
Models, recipes and modules
Term |
Meaning |
|---|---|
Model |
A neural network that predicts objects or tissue classes from images. Its saved parameters are called weights. A module can contain one model or several. |
Recipe |
The module’s instructions: which models to use, their settings, and the order of model and script steps. These are stored in |
Module |
The complete analysis package: its recipe, model weights, and any scripts for processing objects or calculating measurements. |
For example, a testicular analysis could first identify seminiferous tubules, then identify germ cells within those tubules, and finally calculate measurements. The models make predictions; the recipe defines how those steps fit together; the module packages the workflow.
For analysis, choose a trained model suited to your images. Starter models provide a starting point for training, not ready-made predictions for your study.
You do not need to edit a recipe to use an existing module. Select it in Modules and open Workflow to see its steps, output classes and required image resolution. To build or change one, follow Module authoring.
Analysis regions
A region of interest (ROI) is an area selected for analysis. You can draw rectangles with the HistoForge ROI tool or use existing QuPath area annotations through Annotations by class. The latter selects all eligible annotations with the chosen class on each checked image, not just the highlighted annotation. Whole slide analyzes the image without requiring ROIs. See Your first analysis for the region-selection steps.
A tissue region predicted by a model is a result, not an analysis ROI. For example, a model may label areas of epithelium and stroma inside the ROI you drew.
Results and measurements
Segmentation identifies the outlines of individual cells or structures, or the areas occupied by tissue classes. Classification assigns class labels to those results. A tissue-region model does not separate individual cells, so counting its regions is not a cell count.
Results are editable QuPath objects. If the module defines measurements, HistoForge calculates them during analysis. Metrics recalculates measurements from the current objects without rerunning models or processing scripts; Export saves them in an Excel file. Modules without measurements still provide objects for review, but not HistoForge measurement tables or export.
Training and publishing
Fine-tuning adapts an existing model using your annotated images. HistoForge trains one selected model at a time and saves the result as a new module in your Modal workspace. The source module remains unchanged.
You can use that result privately. Publishing is a separate action that copies a module to a Hugging Face repository, where you choose who can access it.