A Nextflow pipeline for analyzing Whole Slide Images (WSI) using GrandQC and foundation models.

This pipeline provides automated analysis of whole slide images using:

• GrandQC for quality control and artifact detection
• Foundation model embeddings for tissue analysis
• TIA Toolbox for image processing and analysis

• Nextflow
• Conda/Mamba
• Python 3.10+
• A HuggingFace token
• Approved access for the relevent model on HuggingFace

1. Clone this repository
2. Create and activate the conda environments:

# Create GrandQC environment
conda env create -f conda/grand_qc.yml

# Create TIA Toolbox environment 
conda env create -f conda/tiatoolbox.yml

Run the pipeline using:

nextflow run main.nf \
    --samplesheet test_data/small.csv \
    -profile conda,local \
    --grand_qc true \
    --foundation true

• samplesheet: Path to the samplesheet
• outdir: Path to the output directory
• grandqc: Should GrandQC be run (default: true)
• foundation: Should the foundation model be run - (default: false)
• foundation_model: Which foundation model should - be used (default: 'H-Optimus-0', other models not - tested)
• huggingface_hub_pat: Path to your HuggingFace cache

• grand_qc.nf - GrandQC workflow
• tia_toolbox.nf - TIA Toolbox processing steps modules - Individual process modules

• grand_qc_run.nf - GrandQC execution
• grand_qc_metrics.nf - Extraction of artefact metrics from GrandQC QC mask
• grand_qc_report - Generate a markdown report from GrandQC outputs
• foundation_model_embedding.nf - Tile embedding with a pathology foundation model
• foundation_model_clustering.nf - Clustering of tiles based on embedding vectors and selection of representative tiles

The pipeline generates:

• Tissue detection masks
• Artifact detection overlays
• QC metrics and reports
• Foundation model embeddings
• Clustering results

The following is the command line output following an example stub run

(base) ataylor@ajt-mbp nf-wsi-vignette % nextflow run main.nf \
    --samplesheet test_data/samplesheet.csv -grand_qc true --foundation true \
    -profile conda,local -stub

 N E X T F L O W   ~  version 24.10.4

Launching `main.nf` [magical_payne] DSL2 - revision: d7073144b0


---------------
NF-WSI-VIGNETTE
---------------
Adam J. Taylor
Sage Bionetworks
---------------
Generating QC reports and feature embeddings for whole slide images
-----------------------------------------------------------------
params:
    samplesheet          : test_data/samplesheet.csv
    outdir               : results
    grandqc              : true
    foundation           : true
    foundation_model     : H-optimus-0
    huggingface_hub_path : /Users/ataylor/.cache/huggingface

profile: conda,local
-----------------------------------------------------------------

executor >  local (55)
[0b/994fd0] process > GRAND_QC:GRAND_QC_RUN (6)        [100%] 9 of 9 ✔
[77/cfc9fc] process > GRAND_QC:GRAND_QC_METRICS (9)    [100%] 9 of 9 ✔
[2a/731da7] process > GRAND_QC:GRAND_QC_METRICS_MERGE  [100%] 1 of 1 ✔
[73/a1fb8c] process > GRAND_QC:GRAND_QC_REPORT (9)     [100%] 9 of 9 ✔
[fe/7528a6] process > TIA_TOOLBOX:EMBEDDING (2)        [100%] 9 of 9 ✔
[8c/c8ef2b] process > TIA_TOOLBOX:CLUSTERING (9)       [100%] 9 of 9 ✔
[24/29345b] process > TIA_TOOLBOX:EMBEDDING_REPORT (9) [100%] 9 of 9 ✔

(grandqc) ataylor@ajt-mbp nf-wsi-vignette %