Technical Overview

CytoCV is a Django-based web application that keeps authenticated research workflows, image-processing stages, segmentation, measurement, and result review together in one browser-based workflow rather than scattering those steps across separate tools.

At a high level, the codebase separates account and workflow concerns from scientific processing concerns, while preserving a run-centered view of uploaded files, derived artifacts, and exported measurements.

The implementation combines HTML template views, request handlers, background workflow coordination, database-backed state, and file-backed run artifacts. Presentation, request handling, persistence, and scientific processing remain distinct layers even though they work together in one application.

Long-running work such as upload preparation and analysis is coordinated outside the initial browser request so validation, preprocessing, segmentation, and quantification can progress through explicit workflow stages.

A typical run begins with supported `.dv`, `.tif`, or `.tiff` upload, channel validation, preview generation, and scale extraction. After that, the workflow advances through preprocess review, segmentation, per-cell measurement, display review, and export.

Each stage keeps the run configuration, channel interpretation, and produced outputs connected so researchers can review what happened at each point in the pipeline instead of only seeing a final spreadsheet.

The default modern analysis path centers on DIC as the structural input for Mask R-CNN-based segmentation. Fluorescence channels then contribute to plugin-scoped measurements such as puncta distance, contour summaries, nuclear or cell-pair intensity, and CEN dot classification.

This separation matters: cell finding is driven by structural context, while fluorescence channels remain the primary source of biological signal quantification. Legacy Blue-based workflows are still supported, but they are not the main default path.

CytoCV persists both database-backed workflow state and run-scoped media artifacts. Uploaded files, preparation jobs, analysis jobs, segmented runs, previews, and per-cell statistics are treated as related pieces of one analysis record rather than isolated files.

The review surfaces then expose outlined images, segmented cell assets, overlays, and table exports so users can move from visual inspection to quantitative output without losing the connection to the run that produced those results.

The major runtime stack combines Django for the web layer, django-allauth for authentication flows, django-tables2 for tabular presentation, NumPy/OpenCV/scikit-image/Pillow for image processing, and TensorFlow/Keras/Mask R-CNN components for inference and segmentation support.

These dependencies are easiest to understand by role: web framework, scientific image processing, inference, persistence, and export. The goal here is to show what each category contributes to the workflow.