Yeast Flowcell Accumulation — PyQt App

PyQt6 desktop app for registration → segmentation → subtraction → evaluation of time-lapse images to quantify yeast attachment over time. Multi-core by default; optional CUDA acceleration when OpenCV with CUDA is available.

Quick start

python -m venv .venv && source .venv/bin/activate  # (Linux/macOS)
# or .venv\Scripts\activate (Windows)

pip install -r requirements.txt
python app/main.py

GPU (optional)

If your OpenCV is CUDA-enabled, the app will automatically use it for some ops (warp, threshold, absdiff). Otherwise it runs on CPU with multi-processing. To install a CUDA wheel, consider prebuilt packages from opencv-python ecosystem or build from source with CUDA. The code checks cv2.cuda presence.

Run Analysis

After selecting an image folder and setting parameters, click Run Analysis to process the time-lapse sequence. Output controls provide options:

• Save diagnostic outputs – write optional masks (gain/loss, overlap/union, segmentation overlays, GM composites).
• Archive and remove images – zip all image subdirectories and delete the originals, keeping summary.csv for results.

Intermediate outputs

The pipeline always writes core results to registered/mov/ and the diff/ subdirectories raw/, bw/, green/, and magenta/. Enabling Save diagnostic outputs adds further artifacts such as diff/new/, diff/lost/, diff/gain/, diff/loss/, diff/overlap/, diff/union/, diff/gm/, and segmentation masks in seg/.

Gain/Loss Detection

After a successful segmentation preview, a collapsible Gain/Loss Detection section becomes available below the segmentation controls. It exposes the thresholding, morphology and saturation settings for detecting newly appeared (magenta) or disappeared (green) regions and includes a preview button. The section and its preview button remain disabled until segmentation completes, and any new folder selection or parameter tweak collapses and disables it again.

The separation between the magenta and green channels is determined in LAB color space using an adaptive threshold on the "a" channel. The composite itself blends the current frame with the previous according to gm_opacity (percentage of the current frame, default 50). A saturation boost (gm_saturation, default 1.0) scales the chroma channel before thresholding to emphasize subtle differences. By default an Otsu threshold (gm_thresh_method="otsu") is used, but a percentile (gm_thresh_percentile, default 99.0) can be selected instead. To remove speckles and recover full structures the masks are optionally processed with morphological closing (gm_close_kernel, default 3) and dilation (gm_dilate_kernel, default 0).

Project layout

• app/main.py — app entry, sets up MainWindow.
• app/ui/main_window.py — PyQt UI and interactions.
• app/models/config.py — dataclasses for parameters; JSON presets; QSettings persistence.
• app/core/io_utils.py — file discovery, timestamp spacing, safe I/O.
• app/core/registration.py — ECC / ORB(+RANSAC) on CPU; CUDA path when available. ECC assumes sufficient texture; nearly uniform frames can cause the optimizer to diverge, yielding NaN transforms that are handled by falling back to the identity matrix.
• app/core/segmentation.py — outline-focused segmentation (black-hat + Otsu, Multi-Otsu, Li, Yen, adaptive, local, or manual thresholding), morphology; skip_outline option bypasses the prefilter for low-contrast images.
• app/core/background.py — on-the-fly background estimation (temporal median of early frames, fallback to blur).
• app/core/processing.py — end-to-end per-frame pipeline; overlap cropping; metrics aggregation.
• app/core/multiproc.py — ProcessPool execution with chunking; CPU core detection.
• app/workers/pipeline_worker.py — background worker orchestration.
• requirements.txt — dependencies.

Low-contrast frames

The segment function accepts a use_clahe flag that applies Contrast Limited Adaptive Histogram Equalization before thresholding. This can recover features in nearly uniform frames where default segmentation yields an empty mask.

from app.core.segmentation import segment

mask_plain = segment(low_contrast_frame)
mask_clahe = segment(low_contrast_frame, use_clahe=True)

With CLAHE enabled (mask_clahe), faint cell boundaries become visible compared to the plain result (mask_plain).