Nuclei Metric Learning

Metric learning for worms nuclei.

Scripts

consolidate_worms_dataset

Call: ./src/scripts/consolidate_worms_dataset -c default.toml -i path_to_30WormsImagesGroundTruthSeg

Creates .hdf datasets for each worm. Gets rid of worm names and mis-matched label numbering by unifying these with universe.txt and worm_names.txt. Most setting are set in .toml config file (check default.toml), such as worms_dataset that here is used as path to output dataset, but throughout the project is used as input path.

HDF keys:

• volumes/raw: raw input, [140x140x1166] uint8, without any modification e.g. normalization
• volumes/nuclei_seghyp: instace labeling, [140x140x1166] uint16, labels no meaning, just a number to distinguish between instances
• matrix/con_seghyp: center of nuclei, each row corresponds to the label in volumes/nuclei_seghyp, [max (nuclei_instances), 3] float32
• volumes/gt_nuclei_labels: ground truth labels, segmentations the same as nuclei_instances, here just the invalid segmentations are removed, and also relabeled according to universe.txt, [140x140x1166] uint16
• matrix/gt_con_labels: same as con_instances but for gt_nuclei_labels, all fixed size of [559x3] float32 , missing labels are np.array([0.0, 0.0, 0.0])

consolidate_cpm_dataset

Call: ./src/scripts/consolidate_cpm_dataset -c default.toml -i path_to_root_kolmogorov_sol_format_both_directions -i2 path_to_nucleinames_corresponding_to_QAP_sols_labeling -i3 path_to_nuclei_name_labels_in_30WormsImageGroundTruthInstanceSeg

Creates cpm dataset, default in ./data/processed (defined in default.toml). .pkl file containing a dictionary with keys '{w1id}-{w2id}' where w1id<w2id and value is a dict of consistent pairwise matchings.

Models

convnet_models: (No Use For Now) a conventional VGG-style network with some conv layers + some fc layers. For extracting embeddings based on patches.

unet: unet model for pixel-wise embeddings.

Project Organization

├── LICENSE
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── experiments        <- keep experiment results
│
├── experiments_cfg    <- config files to reproduce experiments
│
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
│                         the creator's initials, and a short `-` delimited description, e.g.
│                         `1.0-jqp-initial-data-exploration`.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
│
├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
├── src                <- Source code for use in this project.
│   ├── lib            <- useful code for the project
│   │   │
│   │   ├── data       <- data related code, creating dataset code chunks, or typical DL datasets
│   │   │
│   │   ├── modules    <- model definition, together with train, fit, evaluate methods if it requires
│   │   │                   specialized code
│   │   ├── utils      <- general utility functions, summary writing, plotting and visualizations
│   │
│   └── scripts        <- end-point scripts for different tasks
│   │   └── consolidate_dataset.py
│   │   └── train.py
│   │   └── evaluate.py
│   │
│   └── test           <- scripts to help writing code and testing them, not the typical UnitTest but sth similar

---

Project based on the cookiecutter data science project template. #cookiecutterdatascience