Quick Start

The MAPLES-DR dataset is available for download on Figshare. But in the context of training machine learning algorithms we strongly recommend users to install the maples_dr python package to easily download, format and manipulate the dataset.

This page will guide you through the installation process and basic usage of the package: loading the dataset in memory or saving it in a local folder.

Installation

The maples-dr package is available on PyPI and can be installed using pip:

$ pip install maples-dr

Basic Usage

Once imported, MAPLES-DR train or test sets can be loaded in memory with a single line of Python code.

import maples_dr
train_set = maples_dr.load_train_set()
test_set = maples_dr.load_test_set()

If necessary, the dataset archive is automatically downloaded from Figshare, extracted and cached locally. The data is then returned as a maples_dr.Dataset which takes the form of a list of dictionaries containing all MAPLES-DR labels. (For more information, see the Datasets class documentation).

For example, the vessel map of the first sample of the train set can be accessed with:

vessel_map = train_set[0]['vessels']

---

Alternatively, if you’d rather rely on your own code for data loading, MAPLES-DR images can be saved in local folders:

maples_dr.export_train_set('MAPLES-DR/train/')
maples_dr.export_test_set('MAPLES-DR/test/')

As a result of these commands, all the labels of MAPLES-DR are saved as image files in their appropriate folders:

MAPLES-DR/train/
├── bright_uncertains/
│    ├── 20051019_38557_0100_PP.png
│    ├── 20051020_55346_0100_PP.png
│    └── ... (138 image files)
├── cotton_wool_spots/
│    └── ... (138 image files)
├── drusens/
├── exudates/
├── hemorrhages/
├── macula/
├── microaneurysms/
├── neovascularization/
├── optic_cup/
├── optic_disc/
├── red_uncertains/
└── vessels/

MAPLES-DR/test/
├── bright_uncertains/
│    ├── 20051019_38557_0100_PP.png
│    └── ... (60 image files)
├── cotton_wool_spots/
└── ...

Configure the datasets behavior

The dataset behavior can be tailored to ease the integration with your code or specific application. For instance, you might need the images and biomarker maps to have a specific resolution, a specific format (PIL image or numpy array), a specific channel order (rgb or bgr)… The default behavior of the library is configured with the maples_dr.configure() method, and the configuration options are detailed in maples_dr.config.DatasetConfig documentation.

The following example shows how to configure the dataset to return images as numpy arrays (instead of PIL images) and with a resolution of 512x512 pixels:

maples_dr.configure(resize=512, image_format="rgb")

The same method can be used to specify a local path where the library should read MAPLES-DR data, instead of downloading them from Figshare.

maples_dr.configure(
    maples_dr_path="path/to/MAPLES-DR/AdditionalData.zip",
    maples_dr_diagnosis_path="path/to/MAPLES-DR/diagnosis.xls"
)

Finally, a local path to the MESSIDOR dataset can also be specified with this function in order to include the fundus images from MESSIDOR along with the MAPLES-DR labels. (See Using MESSIDOR fundus images for more details.)

maples_dr.configure(messidor_path="path/to/Messidor/directory/")

---

For more information on all the methods presented in this quick start, please refer to MAPLES-DR Quick API documentation.