farabio

farabio is a minimal PyTorch toolkit for out-of-the-box deep learning support in biomedical imaging. out-of-the-box deep learning support in biomedical imaging.

Features:

• Biomedical datasets

• Common DL models

• Flexible trainers (in progress)

Overview

Getting started

How to install

1. Activate conda environment

$ conda create -n myenv python=3.8
$ conda activate myenv

2. Install farabio

A. With pip:

$ pip install farabio -f https://download.pytorch.org/whl/torch_stable.html

B. Setup from source:

$ git clone https://github.com/tuttelikz/farabio.git && cd farabio
$ python -m pip install --upgrade pip setuptools wheel
$ pip install . -f https://download.pytorch.org/whl/torch_stable.html

Reference

data Package

biodatasets Module

biodatasets module provides classes to load public biomedical datasets in a PyTorch friendly manner.

ChestXrayDataset class

class farabio.data.biodatasets.ChestXrayDataset(root: str = '.', download: bool = False, mode: str = 'train', shape: int = 256, transform: Optional[Callable] = None, target_transform: Optional[Callable] = None, show: bool = True)

PyTorch friendly ChestXrayDataset class

Dataset is loaded using Kaggle API. For further information on raw dataset and pneumonia detection, please refer to [1].

References

1(1,2)

https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia

Examples

>>> valid_dataset = ChestXrayDataset(root=_path, download=True, mode="val", show=True)

visualize_batch()

DSB18Dataset class

class farabio.data.biodatasets.DSB18Dataset(root: str = '.', download: bool = False, mode: str = 'train', shape: int = 512, transform: Optional[Callable] = None, target_transform: Optional[Callable] = None, show: bool = True)

PyTorch friendly DSB18Dataset class

Dataset is loaded using Kaggle API. For further information on raw dataset and nuclei segmentation, please refer to [1].

References

1(1,2)

https://www.kaggle.com/c/data-science-bowl-2018/overview

Examples

>>> train_dataset = DSB18Dataset(_path, transform=None, download=False, show=True)

visualize_batch()

HistocancerDataset class

class farabio.data.biodatasets.HistocancerDataset(root: str = '.', mode: str = 'train', transform: Optional[Callable] = None, target_transform: Optional[Callable] = None, download: bool = False, show: bool = True)

PyTorch friendly HistocancerDataset class

Dataset is loaded using Kaggle API. For further information on raw dataset and tumor classification, please refer to [1].

References

1(1,2)

<https://www.kaggle.com/c/histopathologic-cancer-detection/data>`_

Examples

>>> train_dataset = HistocancerDataset(root=".", download=False, mode="train")

visualize_batch()

RANZCRDataset class

class farabio.data.biodatasets.RANZCRDataset(root: str = '.', mode: str = 'train', shape: int = 256, transform: Optional[Callable] = None, target_transform: Optional[Callable] = None, download: bool = False, show: bool = True)

PyTorch friendly RANZCRDataset class

Dataset is loaded using Kaggle API. For further information on raw dataset and catheters presence, please refer to [1].

References

1(1,2)

https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/data

Examples

>>> train_dataset = RANZCRDataset(_path_ranzcr, show=True, shape=512)

visualize_batch()

RetinopathyDataset class

class farabio.data.biodatasets.RetinopathyDataset(root: str = '.', mode: str = 'train', shape: int = 256, transform: Optional[Callable] = None, target_transform: Optional[Callable] = None, download: bool = False, show: bool = True)

PyTorch friendly RetinopathyDataset class

Dataset is loaded using Kaggle API. For further information on raw dataset and blindness detection, please refer to [1].

References

1(1,2)

<https://www.kaggle.com/c/aptos2019-blindness-detection/data>`_

Examples

>>> train_dataset = RetinopathyDataset(".", mode="train", show=True)

visualize_batch()

Handbook

Overview

Deep learning has transformed many aspects of industrial pipelines recently. Scientists involved in biomedical imaging research are also benefiting from the power of AI to tackle complex challenges. Although academic community has widely accepted image processing tools, such as scikit-image, ImageJ, there is still a need for a tool which integrates deep learning into biomedical image analysis. We propose a minimal, but convenient Python package based on PyTorch with biomedical datasets, common deep learning models, and extended by flexible trainers.

What can I do with this package?

• Load public biomedical datasets

• Load common deep learning models

• Do basic image preprocessing and transformations

• Customize training loops to your own needs

Package structure

How to contribute?

You can contribute to this package by reporting issues and/or by sending pull request.

If you find a bug, please report it by opening an issue on Git.

Machine learning for Biomedical Engineers

Activation functions

Binary Step

\[\begin{split}f(x) = \left\{ \begin{array}{lll} 0 & for & x < x_{min} \\ mx+b & for & x_{min} \leq x \leq x_{max} \\ 1 & for & x > x_{max} \end{array} \right.\end{split}\]

\[\begin{split}f'(x) = \left\{ \begin{array}{lll} 0 & for & x \neq 0 \\ ? & for & x = 0 \end{array} \right.\end{split}\]

(Source code, png, hires.png, pdf)

Piecewise Linear

\[\begin{split}f(x) = \left\{ \begin{array}{lll} 0 & for & x < x_{min} \\ mx+b & for & x_{min} \leq x \leq x_{max} \\ 1 & for & x > x_{max} \end{array} \right.\end{split}\]

\[\begin{split}f'(x) = \left\{ \begin{array}{lll} 0 & for & x < x_{min} \\ m & for & x_{min} \leq x \leq x_{max} \\ 0 & for & x > x_{max} \end{array} \right.\end{split}\]

(Source code, png, hires.png, pdf)

Bipolar

\[\begin{split}f(x) = \left\{ \begin{array}{lll} -1 & for & x \leq 0 \\ 1 & for & x > 0 \end{array} \right.\end{split}\]

\[\begin{split}f'(x) = \left\{ \begin{array}{lll} 0 & for & x \neq 0 \\ ? & for & x = 0 \end{array} \right.\end{split}\]

(Source code, png, hires.png, pdf)

Sigmoid

\[f(x)={\frac {1}{1+e^{-x}}}\]

\[f'(x)=f(x)(1-f(x))\]

(Source code, png, hires.png, pdf)

Bipolar Sigmoid

\[f(x)={\frac {1-e^{-x}}{1+e^{-x}}}\]

\[f'(x)={\frac {2e^x}{(e^x+1)^2}}\]

(Source code, png, hires.png, pdf)

Hyperbolic Tangent, TanH

\[f(x)={\frac {2}{1+e^{-2x}}}-1\]

\[f'(x)=1-f(x)^2\]

(Source code, png, hires.png, pdf)

Arctangent, ArcTan

\[f(x)=tan^{-1}(x)\]

\[f'(x)={\frac {1}{1+x^2}}\]

(Source code, png, hires.png, pdf)

Rectified Linear Units, ReLU

\[\begin{split}f(x) = \left\{ \begin{array}{lll} 0 & for & x \leq 0 \\ x & for & x > 0 \end{array} \right.\end{split}\]

\[\begin{split}f'(x) = \left\{ \begin{array}{lll} 0 & for & x \leq 0 \\ 1 & for & x > 0 \end{array} \right.\end{split}\]

(Source code, png, hires.png, pdf)

Leaky Rectified Linear Units, Leaky ReLU

\[\begin{split}f(x) = \left\{ \begin{array}{lll} ax & for & x \leq 0 \\ x & for & x > 0 \end{array} \right.\end{split}\]

\[\begin{split}f'(x) = \left\{ \begin{array}{lll} a & for & x \leq 0 \\ 1 & for & x > 0 \end{array} \right.\end{split}\]

(Source code, png, hires.png, pdf)

Exponential Linear Units, ELU

\[\begin{split}f(x) = \left\{ \begin{array}{lll} a(e^x-1) & for & x \leq 0 \\ x & for & x > 0 \end{array} \right.\end{split}\]

\[\begin{split}f'(x) = \left\{ \begin{array}{lll} f(x)+a & for & x \leq 0 \\ 1 & for & x > 0 \end{array} \right.\end{split}\]

(Source code, png, hires.png, pdf)

SoftPlus

\[f(x)=ln(1+e^x)\]

\[f'(x)={\frac {1}{1+e^{-x}}}\]

(Source code, png, hires.png, pdf)

Changelog

Recent Changes

• fix on spacings by San Askaruly at 2021-08-18 08:56:55

• hide members except visu_batch in biodatasets by San Askaruly at 2021-08-17 18:14:16

• biodatasets images change in docs by San Askaruly at 2021-08-17 17:56:41

Commit

1acf1452f4

Warning

There were uncommitted changes when this was compiled.

FAQ

• How to install pip package to python that uses conda?

  • Solution from Github issue.

• How do i know on which GPU is my model?

  • Solution from Stackoverflow answer.

Indices and tables

• Index

• Module Index

• Search Page