Engines

Every StrataQuest engine follows the same pattern: Input → Processing → Output. Inputs can be raw images, coded images from previous engines, or measurement tables. Processing applies the engine's algorithm with user-configured parameters. Outputs become available for downstream engines.

The engine architecture is organized into functional categories:

• Pre-Processing: Background Removal, Spectral Unmixing, Color Separation, Projection — prepare images before detection
• Detection: Nuclei Detection, Deep Learning Detection, Tissue Detection, Dots Detection, Membrane Detection, Watershed, Total Area, Otsu Threshold — find and segment objects
• Post-Processing: Grow, Remove Objects, Layers, Manual Correction — refine detection results
• Measurement: Standard Measurements, Derived Measurements, Dot Measurements, Membrane Measurements, Raw Data, Density of Events — extract quantitative data
• Classification: Classifier, Assign Classes, Gates, Cutoffs, Phenotypes — categorize objects
• Spatial Analysis: Distance Maps, Proximity Areas, Spatial Phenotyping, Phenotype Interactions — analyze tissue architecture
• Visualization: Scattergram, Histogram, Virtual Channel — display and explore results

The engine as separator: Gonzalez & Woods frame the central problem of image processing as separation — extracting what you want from what you don't. Every StrataQuest engine is, at its core, a separator. Background Removal separates illumination artifacts from biological signal. Nuclei Detection separates cells from background. Spectral Unmixing separates fluorophore contributions from spectral overlap. The Classifier separates cell populations from mixed pools. Even measurement engines separate quantitative features from the raw pixel data they're embedded in.

The signal chain: Pawley describes the image analysis signal chain: specimen preparation → contrast formation → photon collection → digitization → deconvolution → segmentation → measurement. Each link constrains what the next can achieve — you cannot segment what you cannot detect, and you cannot detect what the optics cannot resolve. StrataQuest's engine pipeline mirrors this chain. The ordering is not arbitrary: pre-processing must precede detection, detection must precede measurement, and measurement must precede classification.

Composability: The power of the engine architecture lies in composability. Simple engines combine to solve complex problems. Need to analyze membrane biomarker expression per cell? Chain: Background Removal → Nuclei Detection → Grow (to create cytoplasmic compartments) → Membrane Detection → Standard Measurements on each compartment. Each engine is simple; the composition is powerful.

Reproducibility and the scientific method: The engine pipeline with its explicit parameters constitutes a complete, reproducible analysis protocol. Unlike manual scoring, where "experienced pathologist" is the method description, an engine pipeline can be exactly replicated by any laboratory with the same software. This is not just convenience — it is a requirement for clinical validation studies and regulatory submissions.

A complete multiplex immunofluorescence analysis pipeline for tumor-immune interaction:

1. Pre-Processing: Background Removal on each channel to correct illumination
2. Detection: Nuclei Detection on DAPI to find all cells
3. Compartments: Grow engine to create cytoplasmic rings; Membrane Detection for membrane compartments
4. Measurement: Standard Measurements to quantify mean intensity of CD3, CD8, PD-L1, CK in each compartment
5. Classification: Gates/Cutoffs to define positive/negative thresholds per marker; Phenotypes to assign cell types (Tumor CK+, T-cell CD3+CD8+, etc.)
6. Spatial Analysis: Distance Maps for cell-to-cell distances; Phenotype Interactions to quantify which cell types cluster together; Proximity Areas to define tumor-immune interface zones

Each engine contributes one piece; together they transform a raw fluorescence image into a quantitative map of tumor-immune architecture.