bellasplit 1.1.0

Creating splits of whole slide image scan data for digital pathology deep learning applications

Ballatrix Split

A microservice creating data splits for deep learning applications in digital pathology.

It operates on a storage sqlite database created by the bellastore package.
This database is particularly helpful when querying metadata of specific slides for downstream tasks via our custom digital pathology API bellapi.

Installation

The source code is currently hosted under https://github.com/spang-lab/bellasplit.

Binary installers are available at PyPi.

pip install bellasplit

Usage

Under docs/.env you find a simple template for an .env file that needs to be located in your current environment and hold the path to a yaml config file. A minimal config file is provided under docs/bellameta.yaml. This config in particular defines the valid SPLITNAMEs available to this package.

from dotenv import load_dotenv
load_dotenv()

from bellasplit.types import Splitname
print(Splitname.list())

In order to fetch data for a new split the DataSource interface is to be used, implementing the get_data method.

from bellameta import types as t

from bellasplit.database import Db
from bellasplit.base_split import DataSource, Split, SplitWriter

class Subtyping(DataSource):
    def __init__(self,
                 db: Db):
        super().__init__(db = db,
                            task= t.Task.Subtyping)

    
    def get_data(self, **kwargs):
        # here we have room for fetching and modifying the data
        # in this minimal example we simply call the filter method, i.e.,
        # the fetched data results from searching the data that
        # satisfies the conditions specified in **kwargs
        data = self.filter(**kwargs)
        return data

In order to fetch the data filters can be defined via **kwargs:

db = Db()
source = Subtyping(db=db)

# the conditions to filter for
kwargs = {
    'stain': [t.Stain.HE],
    'class': [t.Subtype.DLBCL, t.Subtype.FL],
    'cohort': [t.Cohort.Example]
}

source_data = source.get_data(**kwargs)

In order to create and write the final split the Split and SplitWriter classes are used:

split = Split(source_data = source_data, splitname = Splitname.from_int(0), test_size = 0.4, val_size=0.2)
writer = SplitWriter(db=db, split=split)
writer.write()

Documentation

Along with the source code, under docs/demo.ipynb, we provide a demo leading you through the features of this package via an application scenario. For further documentation see also the test suite.