jamesdolezal/slideflow repository overview

⁠Overview

Slideflow is a computational pathology Python package which aims to provide an easy and intuitive way of building and testing deep learning models for use in histology image analysis. It is built with both Tensorflow/Keras and PyTorch backends, supporting standard and custom architectures, as well as CLAM. The overarching goal of the package is to provide tools to train and test models on histology slides, apply these models to new slides, and assess performance using analytical tools including predictive heatmaps, mosaic maps, ROCs, and more.

⁠Getting started

Import the module in python and initialize a new project:

import slideflow as sf
P = sf.Project.from_prompt("/path/to/project/directory")

You will be taken through a set of questions to configure your new project. Slideflow projects require an annotations file (CSV) associating patient names to outcome categories and slide names. If desired, a blank file will be created for you when you first setup a new project. Once the project is created, add rows to the annotations file with patient names and outcome categories.

Next, you will be taken through a set of questions to configure your first dataset source. Alternatively, you may manually add a source by calling:

P.add_source(name="NAME",
             slides="/slides/directory",
             roi="/roi/directory",
             tiles="/tiles/directory",
             tfrecords="/tfrecords/directory")

Once your annotations file has been set up and you have a dataset to work with, begin extracting tiles at specified pixel and micron size:

P.extract_tiles(tile_px=299, tile_um=302)

Following tile extraction, configure a set of model parameters:

params = sf.model.ModelParams(tile_px=299,
                              tile_um=302,
                              batch_size=32,
                              model='xception')

...and begin training:

P.train('category1', params=params)

For complete documentation of all pipeline functions and example tutorials, please see the documentation at slideflow.dev⁠.